Anthropogenic Disturbance and Mojave Desert Tortoise (Gopherus Agassizii) Genetic Connectivity

Home , Eldorado Valley, Ivanpah Valley

University of Nevada, Reno

Connecting the Plots: Anthropogenic Disturbance and

Mojave Desert Tortoise (Gopherus agassizii) Genetic Connectivity

A dissertation submitted in partial fulfillment of the

requirements for the degree Doctor of Philosophy in Geography

Kirsten Erika Dutcher

Dr. Jill S. Heaton, Dissertation Advisor

May 2020

THE GRADUATE SCHOOL

We recommend that the dissertation prepared under our supervision by

KIRSTEN ERIKA DUTCHER

entitled Connecting the Plots: Anthropogenic Disturbance and Mojave Desert Tortoise (Gopherus agassizii) Genetic Connectivity

be accepted in partial fulfillment of the requirements for the degree of

DOCTOR OF PHILOSOPHY

Jill S. Heaton, Ph.D. Advisor Kenneth E. Nussear, Ph.D. Committee Member Scott D. Bassett, Ph.D. Committee Member Amy G. Vandergast, Ph.D. Committee Member Marjorie D. Matocq, Ph.D. Graduate School Representative

David W. Zeh, Ph.D., Dean Graduate School

May, 2020

ABSTRACT

Habitat disturbance impedes connectivity for native populations by altering natural movement patterns, significantly increasing the risk of population decline. The Mojave

Desert historically exhibited high ecological connectivity, but human presence has increased recently, as has habitat disturbance. Human land use primarily occurs in

Mojave desert tortoise (Gopherus agassizii) habitat posing risks to the federally threatened species, which has declined as a result. As threats intensify, so does the need to protect tortoise habitat and connectivity. Functional corridors require appropriate habitat amounts and population densities, as individuals may need time to achieve connectivity and find mates. Developments in tortoise habitat have not been well studied, and understanding the relationship between barriers, corridors, population density, and gene flow is an important step towards species recovery.

Genetic tools provide a framework to examine processes like movement and incorporating landscape enhances our understanding of genetic patterns. For tortoises a historically connected landscape coupled with limited dispersal produced a pattern of isolation-by-distance. This dissertation highlighted more recent genetic connectivity by:

(1) assessing population genetic structure and relatedness across a recently disturbed landscape, (2) evaluating the impact of barriers and corridors using simulations of genetic processes, and (3) investigating the relationship between landscape metrics and genetic connectivity using simulations of disturbance scenarios. I genotyped 299 tortoises at 20 microsatellite loci from the Ivanpah Valley region along the California/Nevada border. A ii

fine-scale sampling scheme was applied to evaluate recent gene flow and historical genetic structure. Because the genetic effects of disturbance are often observable after a substantial time lag, I used individually based spatially explicit forward-in-time genetic simulations to test hypotheses related to barriers, population density, and habitat disturbance. I used landscape resistance surfaces representing 17 locations (525 – 625 km2) in southern Nevada and evaluated connectivity success using genetic differentiation.

Three genetic clusters that generally corresponded to valleys and one mountain pass were detected with second order relationships up to 60 km apart, suggesting a greater range of interactions than previously suspected. The correlation between pairwise genetic distances and cost distances revealed reduced genetic connectivity across a railway and a highway bisecting the study area. In simulations of linear barriers, genetic connectivity improved with corridors, but was also influenced by population density.

Low density landscapes experienced reductions in population size and genetic diversity with or without barriers as the result of individuals moving without finding mates and genetic drift. Simulations found that anthropogenic disturbance increased demographic and genetic effects. Disturbed landscapes with high levels of genetic connectivity tended towards low levels of landscape fragmentation with high amounts of suitable habitat.

Urban growth is predicted to exacerbate declines in tortoise populations and genetic connectivity unless intact habitat and populations are adequately protected and connected. This research provides a basis for management actions to protect desert tortoise habitat between existing conservation blocks and restore connectivity along current barriers. iii

ACKNOWLEDGEMENTS

“Look closely at nature. Every species is a masterpiece, exquisitely adapted to the particular environment in which it has survived. Who are we to destroy or even diminish biodiversity?” – E.O. Wilson

This work is dedicated to the Mojave desert tortoise. It was informed, molded, and enthusiastically supported by many individuals, to whom I am grateful. My committee members collectively formed a rigorous scientific team that transformed me from a field biologist into a researcher, and modeled collaboration through their interactions with one another. My major professor, Jill Heaton expertly managed our group, ensuring I never stumbled too far in the weeds. She led me into the fold of academia and believed I could accomplish this goal from our first interview. I am truly fortunate to have been given this opportunity. Kenneth Nussear challenged me to think in ways I never imagined possible and would not tolerate the possibility of defeat. He allowed me the freedom to explore and I cannot imagine reaching this point without his instruction. Amy Vandergast coached me along the path of population genetics, always constructive in her criticism, she helped me expand my knowledge, and is an excellent role model. Marjorie Matocq immersed me in theory and found something positive in my work before gently redirecting me to improve. I am thankful for her every kindness and critique. Scott

Bassett showed genuine enthusiasm for this work and engaged me in diverse perspectives. I also offer my appreciation for Todd Esque, whose collaboration led me to iv

consider wider views, broader implications, and the back story that led us here. He was always willing to discuss, listen, and advocate for this research.

This dissertation began as part of a larger project on conservation corridors for desert tortoises. I would like to thank the principal investigators Ken Nussear, Todd

Esque, and Amy Vandergast for entrusting me with a portion of this important work.

Activities associated with desert tortoise handling were permitted through the U.S. Fish and Wildlife Service (Roy Averill-Murray), Nevada Division of Wildlife (Christy

Klinger), and California Department of Fish and Wildlife (Rebecca Jones). The Bureau of

Land Management (Amy Fesnock and Mark Slaughter), Clark County Desert

Conservation Program (Scott Cambrin and Kimberley Jenkins), and the National Fish and

Wildlife Federation (Eliza Braendel and Anne Butterfield) funded this research, and I thank them for their contributions. The University of Nevada, Reno (UNR) Department of Geography, Graduate Student Association, and Mackay Scholar Award provided academic funding. I appreciate the Department of Geography office staff for assisting with administrative details.

I am immensely grateful to U.S. Geological Survey for field support and invaluable feedback. It was an honor to work with such a dedicated bunch of tireless biologists, led by Todd Esque. Thank you to Felicia Chen for her exceptional organizational skills, attention to detail, willingness to help, and honesty. Kristina Drake provided invaluable professional advice, talking sense to me when I needed it most. I admire Ben Gottsacker for his work ethic, ability to traverse difficult terrain, and humility. I also credit him with turning me into a Green Bay Packers fan even though I do v

not watch football. I appreciate Amanda McDonald for her calm approach and reserved dignity. I always looked forward to working with her. Jordan Swart showed continual enthusiasm, accomplishing the task at hand with a quiet focus and positive outlook.

Special thanks to Sara Murray, not only is she filled with scientific inquiry, she is resourceful and once saved me considerable time and embarrassment when I locked my keys in my vehicle. I am thankful for the efforts of: Patrick Baird, Molly Bechtel, Lesley

DeFalco, Sasha Karosas, Sydney Kelly, Kathy Longshore, Nan Nourn, Jon O’Hearn,

Greg Olson, Megan Rabinowich, and Matt Simes. Field surveys were conducted by Tim

Alvey, Kemp Anderson, Laina Baltic, Mary Baker, Chrystal Bedwell, Kelsi Black, Chris

Blandford, Corey Chan, Lehong Chow, Don Copeland, Gene Drollinger, Chris Fabry,

Kelly Herbinson, Kathryn Hilsinger, Danna Hinderle, Kelly Hunt, Michael Honer,

Audrey Johnson, Chereka Keaton, Kristin Koeper, Colden McClurg, Corey Mitchell,

Jake Mohlman, Freya Reder, Mike Sally, Brian Sandstrom, Chris Scanlan, Kyle Shelp,

Crissy Slaughter, Kathy Simon, Adam Walters, Carrie Warman, Rachel Woodard, and

John Yerger. Additional samples were graciously provided by Kristina Drake, Todd

Esque, and Rachel Woodard. Many thanks to Lee Bice (Clark County Desert

Conservation Program) for providing GIS layers.

I appreciate everyone who instructed me in the technical details and made sure I was up to speed. First and foremost, Anna Mitelberg for her tremendous patience. She is an amazing teacher and trusted friend. Amy Vandergast generously allowed the use of her genetics lab. Mary Peacock and Veronica Kirchoff provided an introduction to extractions. Ken Nussear, Devin Jacobs, Kevin Shoemaker, and Margarete Walden led vi

me through an obstacle course in R, helping me over every hurdle with countless hours of instruction.

I would like to thank my UNR geography cohort, starting with members of the combined Heaton/Nussear labs for much needed moral support. Corey Mitchell is a true friend, with whom I have been able to share many experiences and emotions. I genuinely admire Anjana Parandhaman’s idealism and desire to make the world better. Thank you for your comradery and professional advice. I appreciate time spent commiserating with

Jonathan Deboer, whose positivity gives me perspective. Steve Hromada has taken a leading role in connectivity research and I look forward to future collaborations. I am encouraged by Ally Xiong’s desire to learn, enthusiasm, and words of wisdom. I also acknowledge Stephanie Freund for welcoming me to the lab and Brenden Laurence for his love of weather. The friendship of Heather Benson and Chelsea Cannon during the quiet years gave me a much needed and appreciated sense of community. Victoria

Randlett was instrumental in pushing me to explore diverse viewpoints. Her open door allowed for entertaining debate and I cherish those conversations. Presenting my research to our joint lab group (Heaton/Nussear/Albright) greatly improved its quality and lucidity.

Many people deserve credit for setting me on a path that led to this research. It was first suggested by Linda Allison, who has mentored me through her counsel and was formative in convincing me to believe in myself. Terry Christopher fervently supported and advocated for my professional development and I value his friendship. Thank you to

Kim Field and Lynn Zimmerman for encouragement and advice. vii

Finally, I would like to thank my family. My mother, Rosewitha Dutcher taught me the value of having a purpose outside yourself. Her integrity, kindness, and sense of fun have provided a foundation for my life. My father, James Dutcher instilled in me responsibility and a strong work ethic. Most importantly, he took me outside from an early age, showing me flowers and insects, and talking to me about the importance of conservation for maintaining good trout streams. My husband and best friend, Ben Zyla’s support has been beyond measure. He picked up the pieces when I was too weary, put everything back together again, and has given me all the credit for this accomplishment. He has been my biggest fan and I could not have achieved this without him. viii

TABLE OF CONTENTS

ABSTRACT ...... i

ACKNOWLEDGEMENTS ...... iii

LIST OF TABLES ...... xiii

LIST OF FIGURES ...... xiv

CHAPTER 1. GENES IN SPACE: WHAT MOJAVE DESERT TORTOISE GENETICS

CAN TELL US ABOUT LANDSCAPE CONNECTIVITY ...... 1

ABSTRACT ...... 1

INTRODUCTION ...... 2

MATERIALS AND METHODS ...... 8

Study Area and Sampling ...... 8

Molecular Methods ...... 9

Genetic Diversity ...... 10

Population Structure...... 11

Model Comparisons for Population Structure ...... 13

Relatedness ...... 14

RESULTS ...... 15

Data Quality ...... 15 ix

Genetic Diversity ...... 16

Population Structure...... 16

Model Comparisons for Population Structure ...... 19

Relatedness ...... 19

DISCUSSION ...... 20

Mountain Passes Provide Connectivity Among Valleys ...... 20

Weak but Detectable Population Structure with Gene Flow ...... 21

Genetic Connectivity in Light of Habitat Loss ...... 23

Management Implications ...... 24

ACKNOWLEDGEMENTS ...... 26

REFERENCES ...... 28

TABLES AND FIGURES ...... 39

CHAPTER 2. GENES BY PLACE: WHY ASSESSING MOJAVE DESERT

TORTOISE GENE FLOW WITH LANDSCAPE MODELS IS USEFUL ...... 50

ABSTRACT ...... 50

INTRODUCTION ...... 51

MATERIALS AND METHODS ...... 54

Forward-in-Time Simulation Model ...... 55

Study Landscape and Digital Representation ...... 55 x

Simulation Parameters ...... 56

Genetic Data...... 58

Population Dynamics and Genetic Diversity ...... 58

Population Genetic Structure ...... 59

RESULTS ...... 60

Initial Genetic Data ...... 60

Simulated Landscape Models ...... 61

DISCUSSION ...... 66

Corridors Improve Connectivity ...... 66

Low Population Densities Lose Genetic Connectivity ...... 68

Management Recommendations ...... 68

ACKNOWLEDGEMENTS ...... 70

REFERENCES ...... 71

TABLES AND FIGURES ...... 81

CHAPTER 3. GENES THROUGH TIME: HOW PREDICTIONS OF MOJAVE

DESERT TORTOISE GENETICS CAN INFORM CONNECTIVITY TODAY ...... 92

ABSTRACT ...... 92

INTRODUCTION ...... 93

MATERIALS AND METHODS ...... 96 xi

Study Landscape Resistance Surfaces ...... 97

Modeling Framework and Parameters ...... 99

Population size, Genetic Diversity, and Genetic Structure ...... 100

Quantifying connectivity Success ...... 101

RESULTS ...... 102

Boulder City Conservation Easement North...... 103

Coyote Springs ...... 104

Dry Lake ...... 104

Eldorado Valley ...... 105

Indian Springs ...... 106

Ivanpah Valley ...... 106

Jean/Roach ...... 107

Las Vegas East ...... 108

Las Vegas North ...... 108

Las Vegas West...... 109

Laughlin ...... 110

Mesquite ...... 110

Moapa Valley ...... 111

Red Rock ...... 111 xii

Sandy Valley ...... 112

Searchlight ...... 113

Trout Canyon ...... 113

Predicted Connectivity Success ...... 114

DISCUSSION ...... 117

Disturbance Reduces Gene Flow and Population Size ...... 118

Connectivity Success is Landscape Dependent ...... 119

Management Recommendations ...... 120

ACKNOWLEDGEMENTS ...... 121

REFERENCES ...... 123

TABLES AND FIGURES ...... 135

APPENDIX A: SUPPLEMENTAL MATERIAL FOR CHAPTER 1 ...... 166

APPENDIX B: SUPPLEMENTAL MATERIAL FOR CHAPTER 2 ...... 169

Model Behavior ...... 169

Computational Limitations ...... 170

FIGURES ...... 171

xiii

LIST OF TABLES

CHAPTER 1

Table 1. 1. Genetic diversity statistics for Gopherus agassizii by survey location...... 39

Table 1. 2. Pairwise FST values ...... 40

Table 1. 3. Isolation-by-distance correlation coefficient values ...... 42

Table 1. 4. Maximum Likelihood Population Effects (MLPE) model summaries ...... 44

Table 1. 5. Total number of second order relationships ...... 45

CHAPTER 2

Table 2. 1. Simulation parameters ...... 81

Table 2. 2. Carrying capacities determined by habitat suitability values ...... 82

Table 2. 3. Genetic statistics for genotypes used in simulations ...... 83

CHAPTER 3

Table 3. 1. Conversion factors used to adjust habitat suitability values ...... 135

Table 3. 2. Results for each landscape scenario...... 136

Table 3. 3. Connectivity success index values for current and future disturbance ...... 140

Table 3. 4. Landscape metrics for habitat by neutral and disturbance landscapes ...... 141

Table 3. 5. Landscape metrics of habitat by ability to maintain genetic connectivity .... 144

Table 3. 6. Akaike’s information criterion ranks of the strength of relationships ...... 145

APPENDIX A

Table A 1. Summary of primer and locus information ...... 166

xiv

LIST OF FIGURES

CHAPTER 1

Fig. 1. 1. Map of survey locations centering on the Ivanpah Valley ...... 46

Fig. 1. 2. Cost surfaces used in Maximum Likelihood Population Effects models ...... 47

Fig. 1. 3. STRUCTURE results with sample location used as a prior ...... 48

Fig. 1. 4. sPCA results of the summary of genetic variability and spatial structure ...... 49

CHAPTER 2

Fig. 2. 1. Simulation model landscapes ...... 84

Fig. 2. 2. Map of tortoise locations from a continuous population in Ivanpah Valley ..... 85

Fig. 2. 3. Population genetic structure for genotypes used to seed simulation models .... 86

Fig. 2. 4. Population density model results ...... 88

Fig. 2. 5. Population growth model results ...... 89

Fig. 2. 6. Increased number of loci model results ...... 90

Fig. 2. 7. Heterogeneous landscape model results ...... 91

CHAPTER 3

Fig. 3. 1. Resistance surfaces of study landscape, Clark County, Nevada ...... 146

Fig. 3. 2. 1. Results by each simulated landscape in Clark County, Nevada ...... 147

Fig. 3. 3. Landscape metrics relative to population and genetic statistics ...... 165

APPENDIX B

Fig. B 1. Model behavior results ...... 171

Fig. B 2. Computational limitations model results ...... 172

CHAPTER 1. GENES IN SPACE: WHAT MOJAVE DESERT TORTOISE

GENETICS CAN TELL US ABOUT LANDSCAPE CONNECTIVITY

ABSTRACT

Habitat loss and fragmentation in the Mojave Desert have been increasing, which can create barriers to movement and gene flow in populations of native species. Disturbance and degradation of Mojave desert tortoise habitat includes linear features (e.g. highways, railways, a network of dirt roads), urbanized areas, mining activities, and most recently, utility-scale solar facilities. To evaluate the spatial genetic structure of tortoises in an area experiencing rapid habitat loss, we genotyped 299 tortoises at 20 microsatellite loci from the Ivanpah Valley region along the California/Nevada border. We used a Bayesian clustering analysis to quantify population genetic structure across valley and mountain pass habitats. A spatial principal components analysis was used to further investigate patterns with isolation-by-distance. To explicitly consider landscape features (e.g. habitat and anthropogenic linear barriers), we used maximum likelihood population effects analyses. We quantified recent gene flow through relatedness using a maximum likelihood pedigree approach. We detected three genetic clusters that generally corresponded to valleys separated by mountains, with one genetically distinguishable population in a mountain pass. Pedigree analyses showed second order relationships up to

60 km apart suggesting a greater range of interactions and inter-relatedness than previously suspected. Our results support historical gene flow with isolation-by- 2

resistance and reveal reduced genetic connectivity across two parallel linear features bisecting our study area (a railway and a highway). Our work demonstrates the potential for tortoises to use a range of habitats, spanning valleys to mountain passes, but also indicates habitat fragmentation limits connectivity with relatively rapid genetic consequences.

INTRODUCTION

An important question in conservation ecology is how anthropogenic landscape change impacts movement and population connectivity. Habitat loss and fragmentation can significantly increase the risk of population decline and extinction for native populations by altering natural movement patterns and landscape use (Ewers & Didham 2006;

Haddad et al. 2015; Hand et al. 2014). Integrating genetics with landscape ecology provides a framework to examine the role of heterogeneous habitats in shaping genetic diversity and population structure (Holderegger & Wagner 2008; Manel et al. 2003;

Storfer et al. 2007). Fortunately, the hard-to-observe process of movement of individuals through a landscape can be inferred by examining genetic structure and relatedness

(Dileo & Wagner 2016; Lowe & Allendorf 2010; Slatkin 1985). However, the genetic effects of recent landscape changes that result in habitat loss or linear barriers are often observable only after a substantial lag time that can range from 1-200 generations

(Landguth et al. 2010); therefore, the influence of these changes on genetic diversity may 3

not be observed for several decades in species with delayed breeding and long generation times, such as the Mojave desert tortoise (Gopherus agassizii).

Historically, the Mojave Desert of southern California, southern Nevada, northwestern Arizona, and southern Utah is thought to have exhibited relatively high levels of ecological connectivity (Dickson et al. 2016). For native species like the desert tortoise, which occurs throughout most of this region (Germano et al. 1994; Murphy et al.

2007), highly connected habitat combined with limited individual movement and dispersal have produced a genetic pattern of isolation-by-distance (IBD) with additional differentiation from topographical features (Hagerty & Tracy 2010; Murphy et al. 2007;

Hagerty et al. 2011; Shaffer et al. 2015). Isolation-by-distance is characterized by continuous populations where interbreeding is limited by dispersal distance and distant populations are more genetically differentiated (Wright 1943). The lack of major geographical barriers to movement has resulted in low to moderate levels of genetic differentiation range-wide (pairwise FST 0.011 – 0.132, Hagerty & Tracy 2010), indicative of gene flow occurring in a stepping-stone like pattern (Murphy et al. 2007;

Hagerty & Tracy 2010; Hagerty et al. 2011; Sanchez-Ramirez et al. 2018). This is further supported by radio-telemetry studies of movement and home ranges. For example, a review of Mojave desert tortoise home range size indicates a range from 1–53 ha (median

9.2 ha) with animals capable of traveling 470 – 823 m/day, while males are known to move over 1 km/day (Berish & Medica 2014). However, most daily movements are under 200 m (O’Connor et al. 1994), suggesting that long-distance dispersal primarily occurs sporadically and over multiple generations for this species (USFWS 1994). 4

The historical landscape, characterized by broad interconnected valleys and mountain passes that influenced the population genetic structure and gene flow we measure in desert tortoises today, has changed. Human presence in the North American deserts has increased since the last century, expressed by rapid urban expansion

(Hughson 2009), and a proliferation of vehicular routes from trails to major highways

(Leu et al. 2008), which have caused loss and fragmentation of desert habitat. Rapid urban development, such as within Las Vegas Valley – once a connective region linking tortoise populations across their range (Britten et al. 1997; Hagerty & Tracy 2010), has resulted in substantial loss of habitat connectivity and reduced movement of animals and gene flow relative to historical conditions. Desert valleys along the state line between Las

Vegas, Nevada and the desert cities of southern California have recently undergone substantial habitat alteration. Significant disturbance was initially related to mining throughout the area and has continued to grow since the mid-1800s. The Southern Pacific

Railroad was built in the mid-1880s to support mining and transport people and goods, and still bisects the desert today (Tuma & Sanford 2014). The urbanization of desert lands increased throughout the 1900s and Las Vegas is now a major metropolitan area.

An interstate highway route (I-15) through the Nevada and California Mojave desert can be traced to the early 1900’s, with the interstate we know today largely defined by the mid-20th century. The ever increasing highway speeds and traffic, installation of concrete barriers between north and south bound lanes, and desert tortoise exclosure fencing starting in the late 1990’s along portions of the highway effectively creates a nearly complete barrier to tortoise movement (Peaden et al. 2017). However, the presence of 5

culverts under I-15 may allow for occasional passage as tortoises are known to use storm drain culverts under other highways (Boarman et al. 1997). The large network of dirt roads across the Mojave, although not an observed barrier to tortoise movement, has numerous negative effects on the quality of desert tortoise habitat (as summarized by

Heaton et al. 2008). Most recently, habitat loss has further intensified throughout tortoise habitat with the development of utility-scale solar facilities, which have increased markedly since 2010 (BLM 2010; BrightSource Energy 2014).

Disturbance in arid ecosystems has long lasting impacts (Webb & Wilshire 1980) that may preclude habitat restoration or recovery (USFWS 2011). This poses a serious risk to the long-term persistence of the desert tortoise, which was federally listed as a threatened species under the Endangered Species Act in 1990 largely due to reductions in range and population density (USFWS 1994). Population trends indicate rapid declines associated with human landscape disturbance; specifically, habitat loss and degradation due to urbanization (Averill-Murray et al. 2012; Corn 1994; Doak et al. 1994; Tracy et al.

2004; USFWS 2011). Range-wide, populations have continued to decline since their

1990 listing, reportedly by roughly one-third in the last decade (Allison & McLuckie

2018).

Habitat loss and fragmentation are expected to increase due to ongoing development, which could eventually threaten connectivity for the tortoise (Averill-

Murray et al. 2013). For example, of the 16,282 km2 of tortoise habitat that lies outside conservation areas, 700 km2 has been projected to be lost to utility-scale solar development (Averill-Murray et al. 2013). The timeframe for development is indefinite 6

as projects are being proposed, modified, and constructed continually with no perceivable endpoint. Utility-scale solar placement is also variable and subject to change. As human population growth, urbanization, and utility-scale solar energy construction on public lands continue to significantly reduce habitat for the Mojave desert tortoise (Berry &

Aresco 2014) the likelihood that the species will become reliant on sustained conservation actions increases (Averill-Murray et al. 2012). Given that persistent urban expansion has amplified isolation for tortoises (Averill-Murray et al. 2013), and that development will likely continue, the need to maintain connectivity from California through Nevada and into Utah and Arizona is now more vital than ever. Therefore, understanding existing tortoise population genetic structure is key to assessing the impacts of continued habitat loss and fragmentation.

The area within and surrounding the Ivanpah Valley provides a study region replete with historical and more recent potential anthropogenic barriers to tortoise movement and gene flow, as well as natural features that may either facilitate gene flow

(large areas of open habitat), or restrict gene flow (mountain passes and expansive dry lakes). Because tortoises are commonly associated with desert valleys, but have been recorded in rugged terrain (O’Connor et al. 1994) and are known to occupy and move through heterogeneous habitat (Morafka & Berry 2002), we hypothesized that tortoises have historically used mountain passes as connective habitat between the Ivanpah Valley and adjacent valleys. However, habitat disturbance may alter connectivity.

Anthropogenic barriers within the Ivanpah Valley include I-15 (50–80 years) and the

Southern Pacific Railroad (140 years), while more recent impacts include a golf course (> 7

20 years) and three utility-scale solar developments (< 10 years). The solar installations were sited in previously undeveloped Mojave desert tortoise habitat, where density was estimated between 1.2-10.4 tortoises/km2 (Ironwood Consulting 2012). Developments in valley habitat, including solar energy facilities on public lands, have not been well studied to evaluate impacts to the species (Lovich & Ennen 2011; USFWS 2011), and understanding population genetic structure and gene flow in these areas is vital if genetic connectivity is to be maintained into the future.

Given the relatively long generation time of the desert tortoise (20-25 years;

USFWS 1994) traditional measures of population differentiation (e.g. FST) may not reflect current landscape conditions. Additional analytical methods such as examining the spatial distribution of first and second order relatives can help to understand more recent movement and dispersal patterns (Vandergast et al. 2019). Using clustering approaches and explicitly testing for effects of individual landscape features may help to better characterize the relative impacts of natural and anthropogenic features on genetic structure.

In this study, we applied a fine-scale sampling scheme and combined pedigree reconstruction and genetic clustering analyses with spatially explicit methods to evaluate recent gene flow and historical genetic structure in relation to anthropogenic and historical landscape features within and surrounding the Ivanpah Valley. The specific goals of this research were three-fold: (1) to identify the role of historical landscape features with suitable desert tortoise habitat in facilitating genetic connectivity among adjacent valleys, (2) to assess genetic structure and relatedness across a heterogeneous 8

landscape that has undergone recent and rapid habitat disturbances, and (3) to quantify individual and population level patterns of genetic variation to provide a reference for genetic connectivity for future comparisons. A reference of historical genetic connectivity in this system could be important in understanding the role of intact habitat for tortoise persistence relative to ongoing disturbance.

MATERIALS AND METHODS

STUDY AREA AND SAMPLING

This study was conducted in the central portion of the Mojave desert tortoise range

(Nussear et al. 2009), focusing on the Ivanpah Valley and surrounding Mesquite Valley and Piute Valley connected by mountain passes (Fig.1. 1). Field surveys were conducted in 2015, 2016, and 2017 at ten, 1 km2 study plots (six in Ivanpah Valley, one in Mesquite

Valley CA, one in Eldorado Valley NV, and two in mountain passes between valleys), in a diverse array of suitable habitat. Genetic samples collected from an additional location

(Piute Valley, near Searchlight NV) prior to 2015 were also included, for a total of eleven locations and 299 genetic samples (Fig. 1. 1). The number of individuals varied by study site (Table 1. 1), which may be indicative of natural tortoise densities across the Mojave

Desert. Allison & McLuckie (2018) report adult densities of < 1 – 22.5 tortoises per km2.

Construction at the Ivanpah Solar Electric Generating System (ISEGS; 14 km2) on the west side of Ivanpah Valley, and Silver State Solar (14 km2) on the east began in 2010.

Tortoises from within the ISEGS footprint were translocated to the north of the facility 9

and those from within Silver State Solar were translocated to the east. At our ISEGS

North plot 21 of 53 samples were from translocated animals, and at our Silver State plot

11 of 21 were from translocated animals.

MOLECULAR METHODS

Genetic (blood) samples were collected using subcarapacial venipuncture (Hernandez-

Divers et al. 2002). Samples collected in the field were stored by placing one drop on a

Fast Technology for Analysis (FTA) card (Whatman GE Healthcare Life Sciences); each card was air dried and stored individually in a paper coin envelope. All extractions were performed with the DNeasy Blood and Tissue Kit (Qiagen) using the manufacturer’s instructions, with two minor changes: samples were incubated at 70°C for 10 minutes after the addition of Buffer AL, and the elution step was performed twice with an elution volume of 100 µl for a total final volume of 200 µl. We amplified 20 variable microsatellite loci previously developed for tortoises (Edwards et al. 2003; Hagerty et al.

2008; Schwartz et al. 2003). Amplification of microsatellite loci was performed in 10 µl reactions with 4 µl Multiplex PCR Plus cocktail (Qiagen), 0.8 µl primer mix, 3.2 µl water, and 2 µl DNA diluted to ≤ 4 ng/µl. Thermocycler conditions were set at 95°C for 5 minutes, then 30 cycles were performed with 30 second denaturing at 95°C, 3 minute annealing at 56°C, and 45 second elongation at 72°C, finishing with a 30 minute final elongation at 68°C. PCR product (1 µl) was aliquoted into 10.5 µl HiDi formamide

(Thermofisher) with 0.5 µl LIZ500 (Thermofisher) and submitted to Eton Bioscience

(San Diego, CA) for genotyping. Each round of genotyping included negative controls to 10

check for contamination. Approximately 10% of the samples were amplified and genotyped twice to assess mistyping and dropout rates. We scored raw data in

GENEMARKER v.1.90 (SoftGenetics), binned alleles using R 3.5.3 (R Core Team 2019)

MsatAllele v.1.04 (Alberto 2009), and checked for null alleles with the R package

PopGenReport v.3.0.0 (Adamack & Gruber 2014). Exact tests for Hardy-Weinberg equilibrium and linkage disequilibrium among microsatellite loci were implemented in

GENEPOP v.4.5 (Rousset 2008) with a Bonferroni correction. Microsatellite loci with inconsistent amplification were not included in the dataset.

GENETIC DIVERSITY

We assessed standard measures of genetic diversity for the entire dataset and by survey location. We calculated the number of alleles per locus, observed (Ho) and expected heterozygosity (He), coefficient of inbreeding (F) using adegenet v.2.1.1 (Jombart 2008) in R, allelic richness (Ar, Adamack & Gruber 2014), and mean relatedness coefficients

(rQG, Queller & Goodnight 1989) with 95% confidence intervals in GENALEX 6.5

(Peakall & Smouse 2012). Deviations of Ho from theoretical expectations were evaluated using a Bartlett test for equal variance across microsatellite loci to assess homoscedasticity and a paired t-test to compare the observed and expected population means.

POPULATION STRUCTURE

Genetic structure was evaluated with multiple analytical methods, as well as hierarchically, starting with the entire dataset, then using subsets of the samples based on the genetic clusters detected. We used a Bayesian clustering analysis to infer population structure (STRUCTURE v.2.3.4, Pritchard et al. 2000). We ran the admixture model, which assumes each individual draws some fraction of its genome from each of K population clusters, with correlated allele frequencies, because allele frequencies are expected to be similar for our survey locations. We estimated the probability of K = 1-10 using ten replicate runs of 1,000,000 Markov Chain Monte-Carlo iterations following a burn in of

500,000. We implemented STRUCTURE for the entire dataset with sampling location as a prior, which can improve model output when genetic structure is weak (Hubisz et al.

2009). We calculated the mean log probability of the data (Pr(X|K) in Pritchard et al.

2000), and second order rate of change (ΔK in Evanno et al. 2005). STRUCTURE results were visualized using PopHelper v.2.2.9 in R (Francis 2017).

We also employed a spatial principal components analysis (sPCA) to further investigate cryptic genetic patterns that can result from IBD (Jombart et al. 2008). This multivariate method differs from the previously described STRUCTURE analysis by maximizing the variance in individual allele frequencies while accounting for spatial autocorrelation and assuming no population model (Jombart et al. 2008; Prunier et al.

2014). The genetic patterns found using sPCA were compared to 999 randomized Monte-

Carlo permutations to test whether observed structure differs from the distribution of random expectations. Eigenvalues are generated through Monte-Carlo simulations and 12

represent both genetic diversity (variance) and spatial structure (spatial autocorrelation as measured by global Moran's I). We performed sPCA analyses with the hierarchical approach described above. A spatially explicit connection network of relative neighbors with a genetic distance matrix was created. The product of the variance and the spatial autocorrelation was separated into positive, null, and negative scores representing the magnitude of global (positive) and local (negative) autocorrelation. Global patterns indicate spatial groups or clines, while local patterns detect stronger genetic differences among neighbors than expected among random pairs (Jombart et al. 2008).

We evaluated population genetic differentiation (FST) between survey locations and between inferred genetic clusters. Linearized FST (FST/(1-FST), Rousset 1997) was calculated between survey locations using Analysis of Molecular Variance (AMOVA,

Peakall & Smouse 2012), with an allelic distance matrix and using 999 permutations. We calculated p-values for pairwise comparisons of FST (Jombart 2008) between detected genetic clusters using Weir & Cockerham’s ‘θ’ (1984). We also tested for historical patterns of IBD using Mantel tests (Mantel 1967), with 999 Monte-Carlo permutations, using genetic and geographic Euclidean distance matrices (Jombart 2008) at multiple levels: 1) among all individuals using each tortoise location; 2) grouping individuals by survey location using the mean locations for each plot; 3) among any genetic clusters detected using mean locations for each cluster; 4) within each distinct genetic cluster independently using tortoise location; and 5) within each individual survey plot using tortoise location.

MODEL COMPARISONS FOR POPULATION STRUCTURE

To examine whether recent anthropogenic influences are associated with measurable effects on connectivity among our survey sites, we examined the correlation between pairwise genetic distances using linearized FST and cost distances reflecting the influence of natural and anthropogenic features on resistance to movement. Comparisons by plot location were analyzed using maximum likelihood population effects (MLPE) models in the package ResistanceGA v.4.0-14 (Peterman 2018) in R. Analyses were limited to individuals genotyped in the ten 1 km2 survey sites within and immediately adjacent to the Ivanpah Valley (using genetic data from 275 tortoises), excluding the Piute Valley location due to the disproportionately large distance from other study sites. Cost surfaces hypothesized to have influenced genetic connectivity over longer time periods (e.g. spanning generations) were created representing: 1) Euclidian distance, where a raster was populated with a single value of no resistance to represent IBD; 2) the inverse of modeled desert tortoise habitat (1 - modeled habitat suitability values from Nussear et al.

2009, as in Hagerty et al. 2011) to represent isolation-by-resistance (IBR); 3) the log distance from the interstate rescaled from 0-1; and 4) the log distance from the railway rescaled from 0-1 (Fig. 1. 2). The log distance surfaces were rescaled from 0-1, such that the cost associated with the linear features would not be weighted higher than areas of non-habitat, and the log was used to reflect the more localized effects of these linear features (Fig. 1. 2). While other roads exist in the study area, these are primarily unpaved roads that have accrued more recently. The additional roads exist at a scale that bisects our smaller scale study plots, and while movement may be reduced by these roads (Sadoti 14

et al. 2017), our study animals cross these road networks far more frequently than the larger barriers presented by the interstate highway (no documented crossings) or the railway (two tortoises from the Nipton plot have passed through culverts under the railway once since 2015), which limit the dispersal ability of tortoises (Edwards et al.

2004, Rautsaw et al. 2018).

The relationship of the linearized FST matrix and cost surfaces was analyzed using genetic algorithms to optimize MLPE analyses. We conducted analyses for all single surfaces individually (using the SS_optim function), as well as all combinations of two, and three surfaces, as well as the full model including all four resistance layers (where multiple surface models were analyzed using the MS_optim function, Peterman pers comm.). The models were combined and ranked by Akaike’s information criterion (AIC) to identify the best performing model (Table 1. 4).

RELATEDNESS

To detect the spatial scale of movement over the past 1-2 generations, we used the maximum likelihood pedigree reconstruction implemented in COLONY (Jones & Wang

2010) to estimate first and second order relationships among all individuals, and then mapped the geographic locations of first and second order relative pairs. For COLONY runs, we assumed inbreeding with male and female polygamy, and coded individuals as follows: all adult females (n = 85) as candidates for maternity, all adult males (n = 131) as candidates for paternity, and all individuals as potential offspring. Simulations were run using full-likelihood, medium precision, medium run length, no sibship scaling, and 15

no priors. To avoid excluding pairs based on a single allele, we set the false allele rate to

0.0001. We conducted three independent COLONY analyses varying the random number seed for each. Additionally, the number of runs within each analysis was set to three, increasing the odds of finding the best configurations with maximum likelihood in each

COLONY run. We assigned the expected probability of detecting a mother or father to

25% based on the ratios of newly found tortoises on previously surveyed plots using mark-recapture techniques. Log likelihood plots for each run were examined for convergence. Data from all three runs were consolidated and only pairs found in common in all three runs were considered. Of these, only dyads with ≥ 80% probability of relatedness in each of the three runs were kept (Warner et al. 2016).

RESULTS

DATA QUALITY

We genotyped 299 samples at 20 microsatellite loci. There was indication of null alleles at two loci (GOA4 and GP61). As the evidence for null alleles at these loci was not consistent across multiple sampling locations these loci were retained. All loci conformed to Hardy-Weinberg equilibrium following Bonferroni correction (corrected p-value =

0.003 for α = 0.05), with none consistently in linkage disequilibrium among locations.

The error rate, determined by repeat genotyping, in the 20 microsatellite loci retained was

0.77% and all errors were caused by allelic dropout. Primers and locus information can 16

be found in Appendix A (Supplemental Material), and raw data are available as a USGS

Data Release (https://doi.org/10.5066/P90LIQRI).

GENETIC DIVERSITY

We summarized estimates of genetic diversity and characteristics for the survey locations and overall based on the microsatellite genotype information (Table 1. 1). The number of alleles per locus ranged from 4-43 (푥̅ = 15.9) across loci, and Ar ranged from 5.4-6.3 across survey locations (ISEGS South and ISEGS North; the two plots adjacent to a utility-scale solar facility with previously contiguous habitat). Expected heterozygosity was lowest at ISEGS South (HE = 0.740) and highest at ISEGS North (HE = 0.796). There were significant differences in mean HE between: Mesquite Valley and three survey locations (ISEGS North p-value = 0.030; Stateline Pass p-value = 0.033; Piute Valley p- value = 0.020); ISEGS South and five survey locations (Southpah p-value = 0.020;

Nipton p-value = 0.045; Silver State p-value = 0.048; Sheep p-value = 0.048; Piute

Valley p-value0.005); ISEGS North and McCullough Pass (p-value = 0.004). The inbreeding coefficient was lowest at ISEGS South (F = 0.074) and highest at

McCullough Pass. (F = 0.128). Relatedness rQG ranged from 0.013 at ISEGS North to

0.073 at nearby ISEGS South.

POPULATION STRUCTURE

STRUCTURE results suggested different numbers of genetic clusters depending on the method used to determine the optimal K value (Pr(X|K) = 10 and ΔK = 3; map of 17

assignment to cluster by sampling location Fig. 1. 3A, barplot Fig. 1. 3B). Because

Pr(X|K) may overestimate genetic clusters when there are patterns of IBD, we report ΔK

(except where Pr(X|K) = 1) which may more accurately detect clusters when spatial autocorrelation is present in continuous populations (Evanno et al. 2005; Schwartz &

McKelvey 2009). Individuals from McCullough Pass formed a unique cluster, while

Eldorado Valley and Piute Valley individuals clustered together. The largest genetic cluster was formed by eight of the eleven plots, and included all survey locations in

Ivanpah Valley, Stateline Pass, and Mesquite Valley. Hierarchical analysis of the three main clusters detected additional structure with no strong geographic clustering (Fig. 1.

3C).

Our sPCA results shed additional light on genetic patterns in the study area, finding Ivanpah Valley and Mesquite Valley differing from locations to the east

(McCullough Pass, Eldorado Valley, Piute Valley) using the full dataset (global r =

0.011, p-value = 0.011; Fig. 1. 4A). Hierarchical analysis identified additional structure within Ivanpah Valley along the east and west sides of the valley, roughly corresponding with linear barriers, that include a 140 year old railway, and 60+ year old highway corridor, which parallel one another across much of our study area (global r = 0.009, p- value = 0.031; Fig. 1. 4B). Positive spatial autocorrelation was detected among individuals within the Eldorado/Piute Valley cluster (r = 0.036, p-value = 0.033) and within McCullough Pass (r = 0.049, p-value = 0.001). Evaluation of individuals within

Mesquite Valley and Stateline Pass found no spatial autocorrelation (global r = 0.036, p- value = 0.310; local r = 0.361, p-value = 0.206). 18

Linearized FST values among sampling locations ranged from 0.003-0.040

(ISEGS North to Silver State and Mesquite Valley to Eldorado Valley; Table 1. 2A).

Based on standard permutation across the full dataset genetic differentiation was significant (linearized FST = 0.022, p-value = 0.001). All comparisons with McCullough

Pass and other survey locations were significant, including adjacent plot locations.

Significant genetic structure was also found between Mesquite Valley and all other locations, with the exception of neighboring Stateline Pass. Eldorado and Piute valleys differed significantly from all other plots, except each other, and Piute Valley from one plot in the Ivanpah Valley cluster (Silver State). When locations were combined to correspond to the three main inferred genetic clusters, pairwise FST values ranged from

0.018-0.028, with each comparison statistically significant (Table 1. 2B).

We detected IBD across the study area at all levels tested: among all individuals using each tortoise location, grouping individuals by survey location and using the mean geographic location of all individuals in each survey area, among the three genetic clusters (Ivanpah/Mesquite Valley, McCullough Pass, and Eldorado/Piute Valley) using the mean of individuals associated with each cluster as the geographic location, and within each distinct genetic cluster. Additionally, a correlation between genetic and geographic distances was found within four individual survey plots (Southpah,

McCullough Pass, Eldorado Valley, and Piute Valley; with McCullough Pass also constituting a genetic cluster; Table 1. 3). Of the six survey plots where no IBD was detected, three were recipients of translocated tortoises (ISEGS North and Silver State).

MODEL COMPARISONS FOR POPULATION STRUCTURE

Model comparisons suggested that the distribution of suitable habitat, the Southern

Pacific Railroad, and I-15 were associated with genetic differentiation. The two highest ranking models both included habitat (as the inverse of the habitat suitability model), which effectively indicates IBR, and is consistent with the findings of Hagerty et al.

(2011). Both models also included the Southern Pacific Railroad. These two models differed by their inclusion of I-15, which connects Las Vegas, Nevada and Los Angeles,

California. The interstate parallels the railway for most of our study area, and thus likely represents an additive barrier that would result in similar genetic patterns. The highest ranking model (with the lowest AIC) included habitat distance, the interstate, and the railway (Fig. 1. 2). This model had a weight of 53%, and an R2 value of 0.90 (Table 1. 4).

A second model including only the habitat and the railway had similar performance, with a ΔAIC of only 0.2, a weight of 47%, and an R2 value of 0.90. All other models had lower performance. The full model included all cost surfaces and ranked 3rd, with a ΔAIC >

14.0, and a model weight of approximately 0. Models including habitat performed better than Euclidean distance alone. All models considered ranked well above the null model, where only the intercept was calculated (Table 1. 4).

RELATEDNESS

Evaluating relationship category through pedigree analysis indicated evidence of first order relationships within eight (Stateline Pass, ISEGS North, ISEGS South, Southpah,

Nipton, McCullough Pass, Eldorado Valley, and Piute Valley) of the eleven survey 20

locations. The total number of first order relationships discovered within survey locations was 63. All but one pair of relatives were found within the same 1 km2 plots. There was one first order relationship between a translocated tortoise from ISEGS North and a resident tortoise from ISEGS South. Because they may have been geographically closer prior to facility construction in 2010, this relationship may not necessarily be due to a natural dispersal event. There were 110 unique second order relationships in the survey area, with 59 of these occurring within the same plot and 51 between plots (Table 1. 5), suggesting tortoise movement on multi-generational time scales has occurred throughout the study area. The Euclidean distances between second order relatives found at different survey locations ranged from 5 to 60 km (푥̅ = 22 km).

DISCUSSION

MOUNTAIN PASSES PROVIDE CONNECTIVITY AMONG VALLEYS

The benefits of connectivity include increased exchange of individuals between habitats, with positive impacts on community interactions (Tewksbury et al. 2002) and improved population size and persistence (Henein & Merriam 1990). The Mojave desert tortoise has long been associated with valley bottoms and bajadas, or coalescing alluvial fans

(Germano et al. 1994), and while tortoises are indeed prevalent in these habitats, we also found strong support that mountain passes are occupied by tortoises and provide connectivity between valleys. Connectivity relies on individual or genetic exchange between substantive habitat areas that serve as avenues for dispersal, travel, reproduction, 21

recolonization, and genetic interchange (Beier & Loe 1992). Mountain passes fit this description by allowing gene flow and serving as bridges between neighboring valleys (in comparison with more rugged mountainous areas), and by providing variable levels of connectivity. McCullough Pass shows evidence of admixture with Ivanpah Valley,

Eldorado Valley, and Piute Valley, and all plots share alleles and/or second order relatives, indicating recent gene flow; albeit at lower levels than Stateline Pass. This is possibly due to differences in terrain, available habitat, local population sizes, variation in movement patterns, and/or behavioral differences. Long-term radio-telemetry studies currently being conducted in these areas indicate tortoises are resident in mountain passes, making it likely that these passes also serve as stepping-stone habitats. Mountain pass populations likely contribute to sustained connectivity over generations rather than only intermittent contributions from long distance movements of individuals through passes. Although anthropogenic disturbance is present in the mountain pass locations, it is largely comprised of unpaved roads, which have not presented an absolute barrier to our study animals. However, recent developments on valley floors have the potential to fragment populations. The results of this research provide valuable insights into the significance of mountain passes for supporting landscape connectivity and highlight their importance for conservation planning.

WEAK BUT DETECTABLE POPULATION STRUCTURE WITH GENE FLOW

Bayesian cluster analyses identified three main historical genetic populations: (1)

Ivanpah/Mesquite Valley through Stateline Pass, (2) McCullough Pass, and (3) 22

Eldorado/Piute Valley. Genetic patterns, revealed by sPCA, differed east to west within the Ivanpah Valley. Genetic structure and relatedness within and between valleys indicate that tortoises occupying this central portion of the Mojave Desert do not represent a single panmictic population. Low levels of genetic differentiation within study plots point to recent exchanges of individuals, which is further supported by the presence of second order relationships among study plots, implying that tortoises may require multi- generational dispersal events to maintain connectivity. Our findings reveal weak but detectable genetic structure consistent with the hypothesis that historical habitat was largely continuous and characterized by interconnected valleys, with relatively few restrictive points (e.g., narrow passes between adjacent valleys).

With large areas of connected habitat and fairly contiguous populations, evidence supports historically high levels of gene flow and low levels of genetic differentiation throughout the range of the species (Murphy et al. 2007; Hagerty & Tracy 2010). Our results demonstrate that tortoises in and around the Ivanpah Valley were likely not genetically or geographically isolated in the recent past, and that this could be a regionally important zone for connectivity. This is supported by the work of Hagerty et al. (2011) where a high probability of genetic connectivity was predicted in and around

Ivanpah Valley. These findings make historically connected valleys, like Ivanpah, especially significant for Mojave desert tortoise connectivity given that much of the habitat in the central portion of the range has been lost to the city of Las Vegas, a large urban area where little connective habitat for tortoises remains.

GENETIC CONNECTIVITY IN LIGHT OF HABITAT LOSS

Although it appears that the greater Las Vegas Valley once served as connective habitat in the central portion of the species range (Britten et al. 1997; Hagerty & Tracy 2010;

Nussear et al. 2009), it has been replaced by a large metropolitan area with incumbent infrastructure that represents a barrier to movement and gene flow for desert tortoises.

The urbanization of Las Vegas Valley elevates the need to preserve desert tortoise population connectivity in adjacent valleys, including Ivanpah, which are now more vital than ever as connections for tortoises from California through Nevada and Utah.

However, development pressures extend well beyond the city of Las Vegas to a myriad of human land use practices that result in permanent habitat loss and fragmentation whereby small actions and influences aggregate into larger actions and effects on the landscape that are not explicitly acknowledged; the “tragedy of fragmentation” (Goble

2009).

Although our results support historical as well as relatively recent gene flow for the Mojave desert tortoise, MLPE analysis revealed a reduction in genetic connectivity within the Ivanpah Valley to the east and west of two linear barriers: the Southern Pacific

Railroad and I-15. The signal caused by the railway is likely stronger than that of I-15 because this feature has been on the landscape longer, for approximately 140 years, or around seven tortoise generations. Interstate-15 runs nearly parallel to the railway and has presented a potential barrier to tortoises for 50-80 years, or around four tortoise generations. This could foreshadow increased impacts to Mojave desert tortoises from more recently established barriers and large scale developments in the next several 24

tortoise generations, as the lag time to detect genetic changes is measured in generations.

For example, most utility-scale solar developments have been on the landscape for less than one tortoise generation, but their potential for isolating tortoises can be seen in the amount of habitat that has been lost in areas that once supported contiguous populations.

While the broader long-term impacts of development will likely reduce connectivity for desert tortoises, translocation of animals from development areas may result in subtle genetic effects that appear to increase genetic diversity and admixture initially. However, this signature is expected to be transient. In this research, the ISEGS North and ISEGS

South plots were located on either end of a utility-scale solar facility (ISEGS), and 40% of our samples from ISEGS North were from individuals that were translocated from within the footprint of that facility. This likely contributed to substantial differences between these two plots including: sample size (ISEGS South n = 10; ISEGS North n =

53), the range of genetic diversity (ISEGS South Ar = 5.4; ISEGS North Ar = 6.3), relatedness (ISEGS South rQG = 0.073; ISEGS North rQG = 0.013), and the fact that these plots contained the only first order relationship found among plots.

MANAGEMENT IMPLICATIONS

Mojave desert tortoise recovery actions include protecting existing populations and habitat by conserving intact landscapes and connecting functional habitat (USFWS

2011). However, tortoise habitat continues to be lost with connections between populations at risk from development. Within the Ivanpah Valley recent habitat loss has been caused by a highly traveled interstate, a railway, a network of dirt roads and off 25

highway vehicle tracks, towns, a golf course, mining operations, and several large solar facilities. Unfortunately, we do not have a baseline for the effects of these features on desert tortoise connectivity and gene flow, because genetic studies were not conducted prior to construction (USFWS 2011). Taken together these recent developments have the potential to continue fragmenting habitat and reduce connectivity within current populations. Additionally, our top two MLPE models indicated a significant influence of the Southern Pacific Railroad and/or I-15 on genetic distances, indicating genetic connectivity across these barriers could continue to decrease through time. If connections are sufficiently restricted, we may see increased isolation of the populations residing in

Ivanpah Valley extending to Eldorado Valley, Mesquite Valley, Pahrump, and beyond.

Though not the focus of this study, other valleys in the Mojave Desert, including

Eldorado and Moapa valleys, are experiencing similar habitat loss and fragmentation to that noted in Ivanpah due to roads and solar developments in tortoise habitat, potentially further exacerbating threats to connectivity.

This work provides a basis for determining management actions that could conserve connected tracts of functional habitat between existing blocks of protected land and can be used as a foundation for continued research efforts moving forward.

Prioritizing connectivity corridors for the Mojave desert tortoise across southern

California and southern Nevada through northwestern Arizona, and southeastern Utah for protection could prevent further isolation of populations that are currently connected via suitable undisturbed areas (Dickson et al. 2016). Alternatively, if habitat quality continues to decline and suitable connectivity is lost, population persistence and recovery 26

may necessitate the creation of corridors through barriers (i.e. such as culverts found under some portions of major highways and railways, but see Rautsaw et al. 2018). For tortoises to persist on the landscape a functional ecosystem is required, which has added benefits for other species in the region (Brooks 2000; Averill-Murray et al. 2012). More generally, populations in large and well connected networks are less threatened by extinction (Hanski 1998), therefore, future conservation management plans could benefit by exploring development scenarios that minimize loss of desert tortoise habitat.

ACKNOWLEDGEMENTS

We thank the team of individuals who have contributed to the acquisition of data including Kristina Drake, Felicia Chen, Ben Gottsacker, Amanda McDonald, Jordan

Schwart, Sarah Murray, Steve Hromada, Brett Dickson, and Ironwood Consulting. This work was supported by the U.S. Bureau of Land Management, the U.S. Fish and Wildlife

Service, the National Fish and Wildlife Foundation, and the U.S. Geological Survey

Ecosystems Mission Area Energy and Wildlife Program. We are particularly grateful to

Amy Fesnock (BLM - California) and Mark Slaughter (BLM - Nevada) for their support of our desert tortoise research program. All tortoises were handled according with

USFWS Permit (permit TE-030659-10), Nevada Department of Wildlife Scientific

Collection Permit 317351, University of Nevada, Reno Animal Care and Use Committee protocol (IACUP 00671), and a Memorandum of Understanding with the California

Department of Fish and Wildlife (all to T. Esque). Any use of trade, firm, or product 27

names is for descriptive purposes only and does not imply endorsement by the U.S.

Government.

REFERENCES

Adamack AT, Gruber B (2014) PopGenReport: Simplifying Basic Population Genetic Analyses in R. Methods in Ecology and Evolution 5(4):384–387

Alberto F (2009) MsatAllele 1.0: An R Package to Visualize the Binning of Microsatellite Alleles. Journal of Heredity 100(3):394–397

Allison LJ, McLuckie AM (2018) Population Trends in Mojave Desert Tortoises (Gopherus agassizii). Herpetological Conservation and Biology 13(2):433–452

Anderson D, Burnham K, (2004) Model Selection and Multi-Model Inference, 2ND edn. Springer-Verlag, New York

Averill-Murray RC, Darst CR, Field KJ, Allison LJ (2012) A New Approach to Conservation of the Mojave Desert Tortoise. BioScience 62(10):893–899

Averill-Murray RC, Darst CR, Strout N, Wong M (2013) Conserving Population Linkages for the Mojave Desert Tortoise (Gopherus agassizii). Herpetological Conservation and Biology 8(1):1–15

Beier P, Loe S (1992) A Checklist for Evaluating Impacts to Wildlife Movement Corridors. Wildlife Society Bulletin 20(4):434–440

Berish JE, Medica PA (2014) Home Range and Movements of North American Tortoises. In: Rostal DC, McCoy ED, Mushinsky HR (ed) Biology & Conservation of North American Tortoises, John Hopkins University Press, pp 96–101

Berry KH, Aresco MJ (2014) Threats to Conservation Needs for North American Tortoises. In: Rostal DC, McCoy ED, Mushinsky HR (ed) Biology & Conservation of North American Tortoises, John Hopkins University Press, pp 149–158

BLM (Bureau of Land Management) (2010) Silver State North Solar Project: Clark County, Nevada, USA. URL http://www.blm.gov/style/medialib/blm/nv/field_offices/las_vegas_field_office/energy/n extlight_other/nextlight_map.Par.94625.File.dat/Project%20Fact%20Sheet%2002Nov10. pdf. Accessed November 2015

Boarman WI, Sazaki M, Jennings WB (1997) The Effect of Roads, Barrier Fences, and Culverts on Desert Tortoise Populations in California, USA. In: Proceedings of Conservation, Restoration, and Management of Tortoises and Turtles-An International Conference, pp 149–158

BrightSource Energy (2014) Ivanpah Project Overview. URL http://www.brightsourceenergy.com/ivanpahsolarproject#.U1tk_ldV8F. Accessed November 2015

Britten HB, Riddle BR, Brussard PF, Marlow R, Lee TE Jr (1997) Genetic Delineation of Management Units for the Desert Tortoise, Gopherus agassizii, in Northeastern Mojave Desert. Copeia 1997(3):523–530

Brooks ML (2000) Does Protection of Desert Tortoise Habitat Generate Other Ecological Benefits in the Mojave Desert? USDA Forest Service Proceedings 15(3):68–73

Corn PS (1994) Recent Trends of Desert Tortoise Populations in the Mojave Desert. In: Bury RB, Germano DJ (ed) Biology of North American Tortoises, National Biological Survey, Fish and Wildlife Research 13, pp 85–94 30

Dickson BG, Albano CM, McRae BH, Anderson JJ, Theobald DM, Zachman LJ, Sisk TD, Dombeck MP (2016) Informing Strategic Efforts to Expand and Connect Protected Areas Using a Model of Ecological Flow, with Application to the Western United States. Conservation Letters 10(5) 564–571

DiLeo MF, Wagner HH (2016) A Landscape Ecologist’s Agenda for Landscape Genetics. Current Landscape Ecology Reports 1(3):115–126

Doak D, Kareiva P, Klepetka B (1994) Modeling Population Viability for the Desert Tortoise in the Western Mojave Desert. Ecological Applications 4(3):446–460

Edwards T, Goldberg CS, Kaplan ME, Schwalbe CR (2003) PCR Primers for Microsatellite Loci in the Desert Tortoise (Gopherus agassizii, Testudinidae). Molecular Ecology Notes 3(4):589–591

Edwards T, Schwalbe CR, Swann DE, Goldberg CS (2004) Implications of Anthropogenic Landscape Change on Inter-Population Movements of the Desert Tortoise (Gopherus agassizii). Conservation Genetics 5(4):485-499

Ewers RM, Didham RK (2006) Confounding Factors in the Detection of Species Responses to Habitat Fragmentation. Biological Reviews 81(1):117–142

Evanno G, Regnaut S, Goudet J (2005) Detecting the Number of Clusters of Individuals

Using the Software STRUCTURE: A Simulation Study. Molecular Ecology 14(8):2611– 2620

Francis RM (2017) PopHelper: An R Package and Web App to Analyze and Visualize Population Structure. Molecular Ecology Resoures 17(1):27–32 31

Germano DJ, Bury RB, Esque TC, Fritts TH, Medica PA (1994) Range and Habitats of the Desert Tortoise. In: Bury RB, Germano DJ (ed) Biology of North American Tortoises, National Biological Survey, Fish and Wildlife Research 13, pp 73–84

Goble DD (2009) The Endangered Species Act: What We Talk about When We Talk about Recovery. Natural Resources Journal 49: 1–44

Haddad NM, Brudig LA, Clobert J, Davies KF, Gonzalez A, Holt RD, Lovejoy TE, Sexton JO, Austin MP, Collins CD, Cook WM, Damschen EI, Ewers RM, Foster BL, Jenkins CN, King AJ, Laurance WF, Levey DJ, Margules CR, Melbourne BA, Nicholls AO, Orrock JL, Song D, Townshend JR (2015) Habitat Fragmentation and its Lasting Impact on Earth’s Ecosystems. Science Advances 1(2) e1500052

Hagerty BE, Nussear KE, Esque TC, Tracy CR (2011) Making Molehills out of Mountains: Landscape Genetics of the Mojave Desert Tortoise. Landscape Ecology 26(2):267–280

Hagerty BE, Peacock MM, Kirchoff V, Tracy CR (2008) Polymorphic Microsatellite Markers for the Mojave Desert Tortoise, (Gopherus agassizii). Molecular Ecology Resources 8(5):1149–1151

Hagerty BE, Tracy CR (2010) Defining Population Structure for the Mojave Desert Tortoise. Conservation Genetics 11(5):1795–1807

Hand BK, Cushman SA, Landguth EL, Lucotch J (2014) Assessing Multi-Taxa Sensitivity to the Human Footprint, Habitat Fragmentation and Loss by Exploring Alternative Scenarios of Dispersal Ability and Population Size: A Simple Approach. Biodiversity Conservation 23(11):2761–2779 32

Hanski I (1998) Metapopulation Dynamics. Nature 396(6706):41–49

Heaton JS, Nussear KE, Esque TE, Inman RD, Davenport FM, Leuteritz TE, Medica PA, Stout NW, Burgess PA, Benvenuti L (2008) Spatially Explicit Decision Support for Selecting Translocation Areas for Mojave Desert Tortoises. Biodiversity and Conservation 17(3):575–590

Henein K, Merriam G (1990) The Elements of Connectivity where Corridor Quality is Variable. Landscape Ecology 4(2–3):157–170

Hernandez-Divers SM, Hernandez-Divers SJ, Wyneken J (2002) Angiographic, Anatomic and Clinical Technique Descriptions of a Subcarapacial Veni-Puncture Site for Chelonians. Journal of Herpetological Medicine and Surgery 12(2):32–37

Holderegger R, Wagner HH (2008) Landscape Genetics. BioScience 58(3):199–207

Hubisz MA, Falush D, Stephens M, Pritchard JK (2009) Inferring Weak Population Structure with the Assistance of Sample Group Information. Molecular Ecology Resources 9(5):1322-1332

Hughson DL (2009) Human Population in the Mojave Desert: Resources and Sustainability. In: Webb RH, Fenstermaker LF, Heaton JS, Hughson DL, McDonald EV, Miller DM (ed) The Mojave Desert: Ecosystem Processes and Sustainability, University of Nevada Press, pp 57–77

Ironwood Consulting (2012) Biological Resources Technical Report, Silver State Solar South, Clark County, Nevada. Prepared for Silver State Solar Power South, LLC.

Jombart T, Devillard S, Dufour AB, Pontier D (2008) Revealing Cryptic Spatial Patterns in Genetic Variability by a New Multivariate Method. Heredity 101(1):92-103

Jombart T (2008) adegenet: A R Package for the Multivariate Analysis of Genetic Markers. Bioinformatics 24(11):1403–1405

Jones OR, Wang J (2010) COLONY: A Program for Parentage and Sibship Inference from Multilocus Genotype Data. Molecular Ecology Resources 10(3):551–555

Landguth EL, Cushman SA, Schwartz MK, McKelvey KS, Murphy M, Luikart G (2010) Quantifying the Lag Time to Detect Barriers in Landscape Genetics. Molecular Ecology 19(19):4179–4191

Leu M, Hanser SE, Knick ST (2008) The Human Footprint in the West: A Large-Scale Analysis of Anthropogenic Impacts. Ecological Applications 18(5):1119‐1139

Lovich JE, Ennen JR (2011) Wildlife Conservation and Solar Energy Development in the Desert Southwest, United States. BioScience 61(12):982–992

Lowe WH, Allendorf FW (2010) What can Genetics tell us about Population Connectivity? Molecular Ecology 19(15):3038–3051

Manel S, Schwartz MK, Luikart G, Taberlet P (2003) Landscape Genetics: Combining Landscape Ecology and Population Genetics. Trends in Ecology and Evolution 18(4):189–197

Mantel N (1967) The Detection of Disease Clustering and a Generalized Regression Approach. Cancer Research 27(2):209–220

Morafka DJ, Berry KH (2002) Is Gopherus agassizii a Desert-Adapted, or an Exaptive Opportunist? Implications for Tortoise Conservation. Chelonian Conservation Biology 4(2):263–287

Murphy RW, Berry KH, Edwards T, McLuckie AM (2007) A Genetic Assessment of the Recovery Units for the Mojave Population of the Desert Tortoise, Gopherus agassizii. Chelonian Conservation Biology 6(2):229–251

Nussear KE, Esque TC, Inman RD, Gass L, Thomas KA, Wallace CSA, Blainey JB, Miller DM, Webb RH (2009) Modeling Habitat for the Desert Tortoise (Gopherus agassizii) in the Mojave and Parts of the Sonoran Deserts of California, Nevada, Utah, and Arizona. U.S. Geological Survey Open-File Report 2009-1102

O’Connor MP, Zimmerman LC, Ruby DE, Bulova SJ, Spotila JR (1994) Home Range Size and Movements by Desert Tortoises, Gopherus agassizii, in the Eastern Mojave Desert. Herpetological Monographs 8:60–71

Peaden JM, Nowakowski AJ, Tuberville TD, Buhlmann KA, Todd BD (2017) Effects of Roads and Roadside Fencing on Movements, Space use, and Carapace Temperatures of a Threatened Tortoise. Biological Conservation 214:13–22

Peakall R, Smouse PE (2012) GENALEX 6.5: Genetic Analysis in Excel, Population Genetic Software for Teaching and Research – an Update. Bioinformatics 28(19):2537– 2539

Peterman WE (2018) ResistanceGA: An R Package for the Optimization of Resistance Surfaces using Genetic Algorithms. Methods in Ecology and Evolution 9(6):1638-1647

Pritchard JK, Stephens M, Donnelly P (2000). Inference of Population Structure using Multilocus Genotype Data. Genetics 155(2):945-959

Prunier JG, Kaufman B, Lena JP, Fenet S, Pompanon F, Joly P (2014) A 40-Year-Old Divided Highway does not Prevent Gene Flow in the Alpine Newt Ichthyosaura alpestris. Conservation Genetics 15(2):453-468

Queller DC, Goodnight KF (1989) Estimating Relatedness using Genetic Markers. Evolution 43(2):258–275

R Core Team (2019) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/

Rautsaw RM, Martin SA, Vincent BA, Lanctot K, Bolt MR, Seigel RA, Parkinson CL (2018) Stopped Dead in Their Tracks: The Impact of Railways on Gopher Tortoise (Gopherus polyphemus) Movement and Behavior. Copeia 106(1):135-143

Rousset F (1997) Genetic Differentiation and Estimation of Gene Flow from F-Statistics Under Isolation by Distance. Genetics 145(4):1219-1228

Rousset F (2008) GENEPOP’007: A Complete Re-Implementation of the GENEPOP Software for Windows and Linux. Molecular Ecology Resources 8(1):103–106

Sadoti G, Gray ME, Farnsworth ML, Dickson BG (2017) Discriminating Patterns and Drivers of Multiscale Movement in Herpetofauna: The Dynamic and Changing Environment of the Mojave Desert Tortoise. Ecology and Evolution 7(17):7010-7022

Sanchez-Ramirez S, Rico Y, Berry KH, Edwards T, Karl AE, Henen BT, Murphy RW (2018) Landscape Limits Gene Flow and Drives Population Structure in Agassiz’s Desert Tortoise (Gopherus agassizii). Scientific Reports 8(1):11231

Schwartz MK, McKelvey KS (2009) Why Sampling Scheme Matters: The Effect of Sampling Scheme on Landscape Genetic Results. Conservation Genetics 10(2):10.1007/s10592-008-9622-1

Schwartz TS, Osentoski M, Lamb T Karl SA (2003) Microsatellite Loci for the North American Tortoises (Genus Gopherus) & their Applicability to other Turtle Species. Molecular Ecology Notes 3(2):283–286

Shaffer HB, McCartney-Melstad E, Ralph P, Bradburd G, Lundgren E, Vu J, Hagerty B, Sandmeier F, Weitzman C, Tracy R (2015) Desert Tortoises in the Genomic Age: Population Genetics and Landscape. Draft Final Report to the California Department of Fish and Wildlife

Slatkin M (1985) Gene Flow in Natural Populations. Annual Review of Ecology and Systematics 16(1):393–430

Storfer A, Murphy MA, Evans JS, Goldberg CS, Robinson S, Spear SF, Dezzani R, Delmelle E, Vierling L, Waits LP (2007) Putting the ‘Landscape’ in Landscape Genetics. Heredity 98(3):128–142

Tewksbury JJ, Levey DJ, Haddad NM, Sargent S, Orrock JL, Weldon A, Danielson BJ, Brinkerhoff J, Damschen EI, Townsend P (2002) Corridors affect Plants, Animals, and their Interactions in Fragmented Landscapes. PNAS 99(20):12923–12926

Tracy CR, Averill-Murray R, Boarman WI, Delehanty D, Heaton J, McCoy E, Morafka DJ, Nussear KE, Hagerty B, Medica P (2004) Desert Tortoise Recovery Plan Assessment. U.S. Fish & Wildlife Service, Desert Tortoise Recovery Plan Assessment Committee

Tuma M, Stanford CB (2014) History of Human Interaction with North American Tortoises. In: Rostal DC, McCoy ED, Mushinsky HR (ed) Biology & Conservation of North American Tortoises, John Hopkins University Press, pp 143–148

USFWS (U.S. Fish & Wildlife Service) (1994) Desert Tortoise (Mojave Population) Recovery Plan. U.S. Fish & Wildlife Service, Portland, OR.

USFWS (U.S. Fish & Wildlife Service) (2011) Revised Recovery Plan for the Mojave Population of the Desert Tortoise (Gopherus agassizii). Pacific Southwest Region, Sacramento, CA.

Vandergast AG, Kus BE, Preston KL, Barr KR (2019) Distinguishing Recent Dispersal from Historical Genetic Connectivity in the Coastal California Gnatcatcher. Scientific Reports 9(1):1355

Warner PA, Willis BL, Van Oppen MJH (2016) Sperm Dispersal Distances Estimated by Parentage Analysis in a Brooding Scleractinian Coral. Molecular Ecology 25(6):1398– 1415

Webb RH, Wilshire HG (1980) Recovery of Soils and Vegetation in a Mojave Desert Ghost Town, Nevada, USA. In: Proceedings of 1980 Symposium, A Compilation of Reports and Papers Presented at the Fifth Annual Symposium of the Desert Tortoise Council, 22–24 March, 1980 Riverside, California

Weir BS, Cockerham CC (1984) Estimating F-Statistics for the Analysis of Population Structure. Evolution 38(6):1358–1370

Wright S (1943) Isolation by Distance. Genetics 28(2):114–138 39

TABLES AND FIGURES

Table 1. 1. Genetic diversity statistics for Gopherus agassizii by survey location: sample size (N), allelic richness (Ar), mean observed heterozygosity (HO), mean expected heterozygosity (HE), inbreeding coefficient (F), and relatedness coefficient (rQG) with 95% confidence interval.

Location N Ar HO HE F rQG

Mesquite Valley 12 5.6 0.796 0.757 0.120 0.061 (0.031 – 0.091)

Stateline Pass 25 6.0 0.816 0.790 0.105 0.019 (0.003 – 0.035)

ISEGS North 53 6.3 0.805 0.796 0.115 0.013 (0.007 – 0.020)

ISEGS South 10 5.4 0.838 0.740 0.074 0.073 (0.030 – 0.121)

Southpah 39 6.2 0.790 0.787 0.118 0.031 (0.022 – 0.040)

Nipton 32 6.1 0.811 0.783 0.105 0.033 (0.023 – 0.043)

Silver State 21 6.1 0.786 0.779 0.118 0.021 (0.005 – 0.034)

Sheep 21 6.0 0.788 0.780 0.118 0.037 (0.022 – 0.052)

McCullough Pass 47 5.8 0.772 0.771 0.128 0.060 (0.052 – 0.069)

Eldorado Valley 15 5.7 0.797 0.770 0.126 0.051 (0.027 – 0.076)

Piute Valley 24 6.2 0.781 0.788 0.124 0.036 (0.022 – 0.049)

Overall 299 5.9 0.795 0.811 0.115 -0.003

Table 1. 2. Pairwise FST values (lower) with corresponding p-values (upper). A) Comparisons between survey locations: Mesquite Valley (MV), Stateline Pass (SL), ISEGS North (IN), ISEGS South (IS), Southpah (SP), Nipton (NI), Silver State (SS), Sheep (SH), McCullough Pass (MC), Eldorado Valley (EV), and Piute Valley (PV). B) Comparisons between the three main genetic clusters: Ivanpah/Mesquite Valley (IVMV), McCullough Pass (MC), and Eldorado/Piute Valley (EVPV). Significant p-values after Bonferroni correction are denoted in bold.

A MV SL IN IS SP NI SS SH MC EV PV

MV - 0.092 0.001 0.002 0.001 0.001 0.001 0.005 0.002 0.001 0.001

SL 0.013 - 0.400 0.003 0.006 0.063 0.397 0.014 0.013 0.003 0.012

IN 0.028 0.011 - 0.087 0.569 0.581 0.051 0.029 0.001 0.010 0.017

IS 0.032 0.028 0.018 - 0.020 0.307 0.193 0.005 0.009 0.001 0.001

SP 0.035 0.021 0.007 0.024 - 0.495 0.237 0.002 0.003 0.001 0.001

NI 0.036 0.016 0.012 0.017 0.013 - 0.227 0.009 0.009 0.001 0.015

SS 0.028 0.012 0.003 0.018 0.014 0.015 - 0.123 0.026 0.035 0.101

SH 0.025 0.022 0.019 0.028 0.029 0.025 0.017 - 0.004 0.001 0.004

MC 0.033 0.024 0.025 0.027 0.027 0.025 0.023 0.028 - 0.003 0.001

EV 0.040 0.030 0.027 0.041 0.033 0.034 0.022 0.035 0.033 - 0.517

PV 0.037 0.025 0.027 0.039 0.033 0.027 0.021 0.033 0.035 0.019 -

B IVMV MC EVPV

IVMV - 0.001 0.001

MC 0.018 - 0.001

EVPV 0.018 0.028 -

Table 1. 3. Isolation-by-distance correlation coefficient values (Pearson’s r) using individual geographic locations; by survey location using mean tortoise locations of each survey area; by inferred genetic cluster using mean tortoise locations of each cluster; within each of the three genetic clusters individually: Ivanpah/Mesquite Valley (IVMV), McCullough Pass (MC), and Eldorado/Piute Valley (EVPV); and within each discrete survey location: Mesquite Valley (MV), Stateline Pass (SL), ISEGS North (IN), ISEGS South (IS), Southpah (SP), Nipton (NI), Silver State (SS), Sheep (SH), Eldorado Valley (EV), and Piute Valley (PV).

r DF p-value

By Individual 0.158 298 0.001

By Survey Location 0.450 10 0.027

By Genetic Cluster 0.791 3 0.039

Within IV 0.116 212 0.001

Within MC 0.230 46 0.001

Within EVPV 0.124 38 0.002

Within MV 0.281 11 0.127

Within SL 0.084 24 0.187

Within IN -0.002 52 0.512

Within IS -0.028 9 0.572

Within SP 0.190 38 0.010

Within NI 0.007 31 0.469

Within SS 0.036 20 0.381

Within SH 0.072 20 0.286 43

Within EV 0.204 14 0.040

Within PV 0.134 23 0.040 44

Table 1. 4. Maximum Likelihood Population Effects (MLPE) model summaries for each model considered (Model); Log likelihood (LogLik); the number of parameters (k); resulting Akaike’s information criterion score (AIC); change in AIC from the highest ranking model (ΔAIC); model weights (W) as calculated per Anderson & Burnham (2004); and R2.

Model LogLik. k AIC ΔAIC W R2

Railway + Interstate + Habitat 159.79 10 -311.58 0.00 0.53 0.90

Railway + Habitat 159.68 7 -311.37 0.21 0.47 0.90

Railway + Interstate + Euclidian 152.60 10 -297.21 14.37 0.00 0.87

Railway + Euclidian 152.30 7 -296.59 14.99 0.00 0.86

Railway 152.16 4 -296.33 15.25 0.00 0.86

Interstate + Euclidian 148.95 7 -289.89 21.69 0.00 0.84

Interstate + Habitat 148.13 7 -288.26 23.32 0.00 0.80

Interstate 146.57 4 -285.13 26.45 0.00 0.77

Euclidian 146.48 2 -284.96 26.62 0.00 0.76

Habitat 146.48 4 -284.96 26.62 0.00 0.76

Null 114.68 1 -223.35 88.23 0.00 0.00

Table 1. 5. Total number of second order relationships detected using the program

COLONY among Gopherus agassizii in the Ivanpah Valley area: Mesquite Valley (MV), Stateline Pass (SL), ISEGS North (IN), ISEGS South (IS), Southpah (SP), Nipton (NI), Silver State (SS), Sheep (SH), McCullough Pass (MC), Eldorado Valley (EV), and Piute Valley (PV).

MV SL IN IS SP NI SS SH MC EV PV

MV 1

SL 2 4

IN 0 1 12

IS 0 0 3 0

SP 1 0 4 0 9

NI 0 3 5 1 1 10

SS 1 1 0 0 1 1 0

SH 1 0 0 0 1 0 4 3

MC 0 0 0 0 3 1 1 2 10

EV 0 0 0 0 0 0 0 0 3 5

PV 1 0 2 0 0 4 1 0 0 2 5

Fig. 1. 1. Map of survey locations centering on the Ivanpah Valley, along the California/Nevada border. The 1 km2 plots are indicated by purple squares; the additional sample location is indicated by a purple circle that represents the area from which samples were collected. Developments on the landscape include urban/solar (areas where habitat has been lost and/or fenced to exclude tortoises), major roads/railway (linear barriers to connectivity), minor/dirt roads and mines (representing habitat degradation). 47

Fig. 1. 2. Cost surfaces used in Maximum Likelihood Population Effects (MLPE) models. Top Left: Single value of 0.001 to represent a cost free (Euclidean distance) surface governed only by isolation-by-distance. Top Right: Log of distance to I-15 highway scaled from 0-1. Bottom Left: Log of distance to the Southern Pacific Railroad scaled from 0-1. Bottom Right: Habitat resistance taken as the inverse of modeled desert tortoise habitat from Nussear et al. 2009 (1 - modeled habitat suitability values). 48

Fig. 1. 3. STRUCTURE results with sample location used as a prior. Number of clusters reported using the Evanno method (ΔK), except where the Pritchard method found one cluster (Pr(X|K)). A) Pie charts representing the proportion of each site’s genetic background coming from three main genetic clusters identified prior to hierarchical analysis (ΔK = 3) for the entire study area with sample size ranges from 10 for the smallest pie to 53 for the largest, colors correspond to those found in B. B) STRUCTURE barplot with each vertical bar representing an individual, color scheme matching that found in A and representing the proportion of each individual’s genetic background coming from the three main clusters, with barplots organized generally from the farthest northwest location (Mesquite Valley) through a mountain pass (Stateline Pass) into Ivanpah Valley plots, through a mountain pass (McCullough Pass) to farthest southeast location (Piute Valley). C) Results of hierarchical analysis revealing additional structure (Ivanpah/Mesquite Valley Δ K = 2, McCullough Pass ΔK = 3, Eldorado/Piute Valley ΔK = 2, Pr(X|K) = 1). 49

Fig. 1. 4. sPCA results of the summary of genetic variability and spatial structure, where white and black squares represent the product of the variance and spatial autocorrelation as scores that are positioned by spatial coordinates. Square size indicates the magnitude of the variance. Negative scores represent local patterns of spatial autocorrelation and positive scores represent global patterns. A) Map of the entire study area showing a pattern of genetic variability between Ivanpah Valley and Mesquite Valley and locations to the east, with potential integration at the Sheep and Nipton plots; and B) Map of Ivanpah Valley revealing an east-west genetic pattern. 50

CHAPTER 2. GENES BY PLACE: WHY ASSESSING MOJAVE DESERT

TORTOISE GENE FLOW WITH LANDSCAPE MODELS IS USEFUL

ABSTRACT

Anthropogenic linear features (road and railways) can fragment Mojave desert tortoise

(Gopherus agassizii) populations. Attention to genetic connectivity in light of disturbance is critical in working towards species recovery. Culverts, often constructed for drainage, may restore genetic connectivity by allowing for safe passage across anthropogenic barriers. However, population density is an additional consideration in connectivity planning that has been given little attention in desert tortoises. I used individually based spatially explicit genetic simulations of gene flow in landscape scenarios constructed as resistance surfaces of (1) neutral landscapes - without barriers and allowing for isolation- by-distance, (2) with a semi-permeability (representing culverts) along linear barriers, and (3) an absolute barrier to dispersal. Simulations were run for 200 non-overlapping generations using 20 variable microsatellite loci derived from an empirical dataset. To evaluate average outcomes, I performed 30 replicate runs. I examined the total number of individuals, genetic diversity (A, Ho), and population genetic structure (pairwise FST,

STRUCTURE, and sPCA) for simulations of population density (low, moderate, high), population growth rate (low), increased number of loci (80), and a heterogeneous landscape. I found that relative to the absolute barrier connectivity improved when corridors allowed for movement across linear barriers and with higher tortoise densities. 51

Population genetic structure increased as barriers reduced connectivity regardless of the modeling scenario. However, low population density simulations lost over 75% of their population size regardless of the landscape scenario. Genetic diversity in the no disturbance landscape simulations was initially high in low population density models (A

= 23.3 ± 0.18, Ho = 0.790 ± 0.004), but fell through time (A =12.8 ± 0.50, Ho = 0.31 ±

0.021), with no significant differences between the three landscapes. My results predict that the addition of connectivity culverts to major roads and railways will improve genetic connectivity for the desert tortoise; however, the consequences of low density may result in greater risk of genetic drift and harmful stochastic demographic processes.

Therefore, care should be taken to ensure Mojave desert tortoise populations remain connected without losing population density.

INTRODUCTION

The degree to which a landscape facilitates movement among species habitat areas is a key metric influencing biodiversity, viable population sizes, the potential for demographic rescue, movements in response to environmental change, gene flow, and genetic resilience across a broad range of taxa (Christie & Knowles 2015; Haddad et al.

2003; Taylor et al. 1993; Tewksbury et al. 2002). Anthropogenic linear features, like roadways, have been found to negatively impact population genetic diversity and genetic structure across taxa by restricting movement (Balkenhol & Waits 2009; Holderegger &

Di Giulio 2010). Increased mortality and modified animal behavior has been documented 52

in multiple species (Trombulak & Frissell 2000). The Mojave desert tortoise (Gopherus agassizii) is effected by direct mortality when roadways are unfenced, or an inability to cross when fenced (Boarman et al. 1997). Roadways have been found to reduce movements of desert tortoises, both in terms of crossing and home-range size (Peaden et al. 2017). Relative abundance is also reduced ≥ 0.2 – 4 km from roadways (Boarman &

Sazaki 2006; vonSekendorff Hoff & Marlow 2002; Nafus et al. 2013).

When a native landscape becomes fragmented, corridors may maintain or even restore connectivity (Noss 1987). Functional corridors allow species to move through habitat embedded in a suboptimal matrix and enhance population viability (Henein &

Merriam 1990; Beier & Noss 1998; Beier et al. 2008). Rautsaw et al. (2018) found that railways restrict movement patterns in gopher tortoises (Gopherus polyphemus), but trenches dug underneath the rails allow for crossing. The strategic placement of underground culverts may optimize safe passage and improve connectivity for Mojave desert tortoises (Boarman et al. 1997; Peaden et al. 2017). Therefore, evaluation of the effectiveness of these corridors is of critical importance (Gregory & Beier 2014).

Fragmented populations are at greater risk of population decline and genetic drift

(Dixo et al. 2009; Holderegger & Di Giulio 2010). The addition of connectivity routes onto the landscape may be insufficient to maintain genetic connectivity if individuals have to travel long distances without finding mates (Bruford et al. 2010). Population decline has been given little attention in Mojave desert tortoise connectivity research. The species was federally listed as threatened due to range-wide declines in habitat and population density (USFWS 1994), and since listing has continued to decline (Doak et al. 53

1994; Corn 1994; Tracy et al. 2004, USFWS 2011), with recent numbers as low as 37% range-wide in the decade from 2004-2014 (Allison & McLuckie 2018).

The cost of movement, reduction in survival, or willingness of an individual to move through its environment can be represented using a landscape resistance model

(Zeller et al 2012), and movement depicted as a function of features on a map using resistance values (i.e. high resistance values may be assigned to urban areas or major roads) for each pixel cell in a gridded raster (Cushman et al. 2013). The sampling grain

(size of the sampling unit; i.e. raster pixel size) should ideally be smaller than an average home-range size or dispersal distance (Anderson et al. 2010). The mean home-range for desert tortoises is highly variable (1.3-53 ha, Berish & Medica 2014) and maximum movements have been recorded between 0.5 - 1.6 km (Nussear et al. 2012).

In species with limited dispersal, such as the desert tortoise, genetic tools can provide a framework to examine hard to observe processes (Brooks 2003; Cushman et al.

2013; Lowe & Allendorf 2010; Dileo & Wagner 2016; Slatkin 1985). Incorporating landscape structure into analyses of movement enhances our understanding of the role landscape features play in shaping genetic diversity and population structure

(Holderegger & Wagner 2008; Manel et al. 2003; Sork & Waits 2010; Storfer et al.

2007).

Evaluations of corridor effectiveness can be challenging because of the lag time associated with detection of genetic patterns following corridor installation (Gregory &

Beier 2014). In these types of analyses, time is best measured using the generation time for the species of interest, rather than number of years. In spatially explicit, individually- 54

based simulations time to detect genetic signals as the result of a landscape barrier ranged from 1 to 200 generations, depending on dispersal ability and genetic response metric

(Landguth et al. 2010). Because desert tortoises have long generation times (20-25 years,

USFWS 1994) it would likely require decades to empirically determine the genetic effects of fragmentation from linear barriers and potential connectivity value of underground culverts. However, forward-in-time simulations may aid in evaluating how changing landscape features are predicted to influence future population genetic patterns

(Epperson et al. 2010; Rebaudo et al. 2013; Rebaudo et al. 2014).

Understanding the relationship between gene flow, population density, connectivity corridors, and physical barriers is important for desert tortoise recovery

(USFWS 2011). Little work has been done to determine how linear barriers may impede connectivity in otherwise connected habitat or evaluate the influence of connectivity culverts on genetic architecture in desert tortoises. This study uses forward-in-time simulation modeling to predict tortoise genetic connectivity in simulated landscapes without a linear barrier, with a semi-permeable barrier, and with an absolute barrier to dispersal to test specific hypothesis. I hypothesized that adding permeability onto a linear barrier will improve genetic connectivity, but that connectivity may decrease with reductions in population density.

MATERIALS AND METHODS 55

FORWARD-IN-TIME SIMULATION MODEL

I preformed individually-based spatially explicit genetic simulations of gene flow across

200 generations in the landscape scenarios using the program SIMADAPT v.1.8.0

(Rebaudo 2014). SIMADAPT uses the NETLOGO environment (Wilensky 1999) to model mating and dispersal in non-overlapping generations using a georeferenced area with closed boundaries and three landscape characterization files: habitat type (here used as a proxy for geographic populations by assigning individuals to groups relative to a geographic barrier), carrying capacity of each grid cell, and landscape resistance. The model simulates landscape genetic processes with user defined simulation parameters, including initial genetic structure, and records the alleles of all individuals from forward- in-time generations (Rebaudo et al. 2013).

STUDY LANDSCAPE AND DIGITAL REPRESENTATION

Landscapes were constructed and mapped in R 3.5.3 (R Core Team 2019) using packages ggmap v.3.0.0.901, raster v.2.9-5, and rgeos v.0.4-3 (Kahle & Wickman 2013; Hijmans

2019; Bivand & Rundel 2019). The landscapes were hypothetical areas ≥ 25 x 25 km of 1 km2 grid cells, constructed as (1) neutral landscapes - without barriers and allowing for isolation-by-distance (IBD), (2) with a semi-permeable linear barrier, and (3) an absolute barrier to dispersal. I used resistance values of 0 (no resistance), 0-0.6 (variable resistance), 1 (absolute barrier), and 0.7 (permeable areas embedded within a barrier; Fig.

2. 1). Values for variable resistance and permeability were derived from the habitat 56

suitability model and expert opinion as being within the range of values appropriate for tortoise occupancy and movement (Nussear et al. 2009). Models with no resistance, barriers, and permeability along barriers allowed for evaluation of differences in genetic diversity and structure in simplified scenarios. Randomly generated variable resistance values were applied to heterogeneous landscape models to incorporate more realistic landscapes.

SIMULATION PARAMETERS

Simulations were run to generate genotype files and investigate six model scenarios: model behavior, computational limitations, population density, population growth, increased number of loci, and heterogeneous landscape (Table 2. 1). Model behavior used simplified landscape scenarios (neutral landscape with no resistance and absolute barriers) to verify models conformed as predicted prior to modeling scenarios with unknown outcomes. Computational limitations tested for unreliability in run completion as a direct result of simulated landscape size and number of individuals to ensure models would be unencumbered by computing limits. Further details on model behavior and computational limitations can be found in Appendix B (Supplemental Material).

In all simulations, individuals were allowed to move up to ten grid cells per generation (≤14 km). As there are no empirical data on an individual’s dispersal likelihood, I used telemetry data from Ivanpah Valley to calculate the percent of animals that left 1 km2 study plots without returning to approximate probability of dispersal at

50% (Hromada pers comm.). To account for uncertainty in population growth rates I ran 57

models using a low of 0.5% annual growth (Turner 1986) and moderate estimate of 1%

(USFWS 1994). Both were multiplied by 48 breeding years, based on average lifespan

(USFWS 1994; Medica et al. 2012) minus average age of reproductive maturity (17 years, McCoy et al. 2014; USFWS 2011). Annual density estimates for desert tortoises are highly variable (0.2-28/km2, Allison & McLuckie 2018). Based on tortoise densities in 1 km2 plots in a current study in Ivanpah Valley, along the Nevada/California border, I calculated 24/km2, and used this value for high density in these models. Medium and low densities were estimated at 14/km2 and 3/km2. Simulations in heterogeneous landscapes were initiated with Mojave desert tortoise genotypes at landscape carrying capacity

(maximum density supported by the habitat) based on the relationship between habitat suitability and population density (Table 2. 2; Nussear et al. in prep).

I simulated gene flow for 200 non-overlapping generations, sampling individuals every five generations. I used a mutation rate of 0.0005 per locus per generation (Dileo et al. 2013; Estoup & Angers 1998; Landguth et al. 2010). This value falls within the range of mutation rates for microsatellite loci estimated by Edwards et al. (2015) for desert tortoise microsatellites. Because increasing the number of loci can decrease sampling error and increase sensitivity (Meirmans 2015) I ran simulations with 20 and 80 loci to determine if changes in genetic diversity or population genetic structure could be detected earlier in time or more clearly with more loci. Simulation parameters by model type are summarized in Table 2. 1.

GENETIC DATA

Neutral genetic markers are not influenced by selective forces, making them ideal for investigations of gene flow (Holderegger et al. 2006). I started with frequencies observed in an empirical dataset of 20 variable microsatellite loci (Edwards et al. 2003; Hagerty et al. 2008; Schwartz et al. 2003) sampled from a continuous population in the Ivanpah

Valley (Fig. 2. 2; Dutcher et al. 2020). Statistics from these data were used to parameterize genotypes for simulations models with increased number of loci (n = 80). I randomized the samples to remove any potential signal of IBD in order to create seed genotypes. These genotypes were simulated forward-in-time with no landscape resistance, using a burn-in of 100 generations to create a large genotype file from which to subsample as input for models of population density, population growth, and heterogeneous landscapes. I tested for departures in Hardy-Weinberg equilibrium, applying a Bonferroni correction, and examined genetic diversity and population genetic structure in the original data from Ivanpah Valley and the simulated data. Simulations were seeded with genotypes equal to the number of individuals at carrying capacity in neutral landscapes.

POPULATION DYNAMICS AND GENETIC DIVERSITY

To account for stochasticity in simulations and ensure understanding of average outcomes I performed 30 repetitions of modeled scenarios sampling genotypes every five generations. I compared population dynamics in neutral landscape models with barrier models, by evaluating the total number of individuals through time. In simulated datasets 59

with > 750 individuals I randomly sampled without replacement to create subsamples for analyses. Evolutionary potential was calculated using genetic diversity statistics as the number of alleles/locus (A) and observed heterozygosity (Ho) using the R package adegenet v.2.1.1 (Jombart 2008). Simulated data were examined through time, using outcomes averaged across replicate simulations.

POPULATION GENETIC STRUCTURE

I investigated population genetic structure in the original genotypes, following randomization, and simulation output using pairwise genetic differentiation (FST, Nei

1973) in the R package hierfstat v.0.04-22 (Goudet 2005), spatial principal components analysis (sPCA) using the R package adegenet v.2.1.1 (Jombart et al. 2008), and a

Bayesian clustering analysis (STRUCTURE v.2.3.4, Pritchard et al. 2000). In simulated data

I evaluated FST in time-series, using outcomes averaged across replicate runs. I also examined population genetic structure with genotype files best representing the mean FST at generation 200 to ensure capture of the effects of landscape. STRUCTURE analyses were performed using the admixture model, with correlated allele frequencies, and location as a prior, which improves inference when genetic structure is weak. I estimated the probability of K population clusters = 1-10 using ten replicate runs of 1,000,000 Markov

Chain Monte-Carlo iterations following a burn-in of 500,000. I calculated the mean log probability of the data (Pr(X|K) in Pritchard et al. 2000). Because Pr(X|K) may overestimate genetic clusters when there are patterns of IBD I also calculated the second order rate of change (ΔK in Evanno et al. 2005). Results were visualized using 60

PopHelper v.2.2.9 in R (Francis 2017). Because STRUCTURE may misrepresent genetic clustering when spatial autocorrelation is present (Pritchard et al 2010; Schwartz &

McKelvey 2008) I used sPCA to evaluate cryptic genetic patterns in the presence of IBD.

This multivariate method differs from STRUCTURE by maximizing genetic diversity

(variance) in individual allele frequencies while accounting for spatial structure (spatial autocorrelation measured by Moran's I). The genetic patterns were compared to 999 randomized Monte-Carlo permutations to test for differences between observed structure and the distribution of random expectations.

RESULTS

INITIAL GENETIC DATA

The simulated dataset produced a large number of individuals (N = 14572); therefore, I used a subsample (n = 750) for comparison with the original genetic data (n = 170). All loci in the original dataset and the subsample conformed to Hardy-Weinberg equilibrium following Bonferroni correction (p < 0.003). The number of alleles ranged from 4 – 39 in the original genetic dataset to 8 – 54 in the subsample. In the original dataset Ho did not deviate from theoretical expectations (p = 0.429, df = 19). Comparisons of the original dataset and the subsample of simulated data found no significant differences in observed and expected heterozygosity (p = 0.369 and 0.038; respectively, df = 19; Table 2. 3).

In the original data no spatial autocorrelation was found using sPCA. However,

STRUCTURE did detect genetic clusters (Pr(X|K) = 5, ΔK = 2) with admixture. In the 61

simulated data, sPCA identified IBD, which is expected in desert tortoises as the result of simulating genotypes across a large landscape surface (625 km2). STRUCTURE found evidence for a continuous population with strong admixture that aligned well with sPCA results (Pr(X|K) = 1, ΔK = 3; Fig. 2. 3).

SIMULATED LANDSCAPE MODELS

I examined the total number of individuals, genetic diversity (A, Ho), and genetic differentiation for models of population density (low, moderate, high), population growth rate (low), increased number of loci (80), and heterogeneous landscape. I compared neutral landscapes with barrier models (absolute and semi-permeable) for each. I found that the number of individuals was highest in neutral landscapes and lowest with an absolute barrier in all models. Genetic diversity followed the same pattern. Genetic differentiation was always lowest in neutral landscapes. Absolute barriers created isolated populations while semi-permeable barriers allowed for some admixture. I was able to detect population genetic structure earlier in time in models with low population density and increased number of loci.

Population density was assigned as 3/km2 (low), 14/ km2 (moderate), and 20/km2

(high). The low density models were seeded with N = 1875 and lost over 75% of the population by generation 200 regardless of landscape (neutral N = 460 ± 12.34, semi- permeable barrier N = 438 ± 8.09, absolute barrier N = 434 ± 9.73). Moderate density models were seeded with 8750 individuals, and remained stable through time.

Populations in semi-permeable barrier landscapes (N = 8034 ± 15.58 at generation 200) 62

and absolute barrier landscapes (N = 8024 ± 14.05) decreased initially, but stabilized within 5 – 10 generations. The population decreased in the barrier model 8.3% by generation 200. Populations in high density models started with 12500 individuals and remained stable in neutral landscapes. Populations initially decreased in semi-permeable barrier landscapes (11540 ± 22.70 at generation 200) and absolute barrier landscapes

(11531 ± 19.65), but stabilized within 5 – 10 generations. The absolute barrier model exhibited a loss in population of 7.5%.

For population density models, in neutral landscapes, the mean number of alleles/locus was initially comparable with low (A = 23.3 ± 0.18), moderate (A = 23.3 ±

0.20), and high density (A = 23.3 ± 0.25). By generation 200 it decreased at low densities

(A =12.8 ± 0.50, p = 1 * 10-5), and there were no significant differences in the three low density scenarios (p > 0.05, df = 19). Low density models also lost heterozygosity in the neutral landscape between generation 0 (Ho = 0.790 ± 0.004) and generation 200 (Ho =

0.31 ± 0.021; p = 3 * 10-15, df = 19). Greater losses were predicted at low density, with semi-permeable (Ho = 0.305 ± 0.016) and absolute barrier models (Ho = 0.287 ± 0.0168), but were not significantly different from the neutral model (p > 0.05, df = 19). By generation 200 the mean number of alleles/locus was relatively comparable to generation

0 in the moderate density model in the three landscape scenarios (p > 0.05, df = 19).

Heterozygosity in moderate density models was not significantly different through time or by landscape scenario (p > 0.05, df = 19). The high density model increased in number of alleles/locus in the neutral models between generations 0 and 200 (A = 24.6 ± 0.42, p >

0.008, df = 19); however, there were no significant differences by landscape scenario (p > 63

0.05, df = 200). Heterozygosity in the high density models was not significantly different through time or by landscape scenario (p > 0.05, df = 19).

Genetic differentiation in population density models was < 0.002 at generation 0 in all simulations, with neutral landscapes displaying stability. Genetic differentiation increased through time in absolute barrier models, regardless of population density.

Absolute barrier landscapes showed the highest genetic differentiation at generation 200 at low density (FST = 0.073 ± 0.008) followed by moderate (FST = 0.012 ± 0.002) then high density (FST = 0.002 ± < 0.001). Genetic differentiation values for semi-permeable barrier landscapes fell between neutral and absolute barrier landscapes at all modeled densities, as expected. Population genetic structure at generation five was absent in all modeled landscapes at all population densities. By generation 200 the neutral landscape revealed spatial autocorrelation at all densities as the result of individual movement abilities relative to landscape size, with no apparent geographic population structure (low density Pr(X|K) =10, ΔK = 3; moderate Pr(X|K) = 1, ΔK = 6; high Pr(X|K) = 1, ΔK = 2).

Absolute barriers resulted in isolation, regardless of density (low and moderate density

Pr(X|K) = 10, ΔK = 2; high Pr(X|K) = 6, ΔK = 2). Semi-permeable landscapes exhibited population genetic structure on either side of the linear barrier; however, with evidence of admixture (low density Pr(X|K) = 10, ΔK = 2; moderate and high Pr(X|K) and ΔK = 2;

Fig. 2. 4).

Moderate population density (N = 8750) was used to initiate all models at a low population growth rate (r = 0.24). The number of individuals was comparable to those found in moderate population density models (r = 0.48). The population decreased by 7.3 64

– 7.4% by generation 200 in the semi-permeable and absolute barrier landscapes. In all landscapes modeled, and through time, the mean number of alleles/locus ranged from

23.0 ± 0.4 – 23.5 ± 0.4 (p > 0.05, df = 19). Heterozygosity remained consistent in all landscapes and by generation, with the high at generation 200 in the neutral landscape

(Ho = 0.792 ± 0.007) and the low at generation 200 in the absolute barrier landscape (Ho

= 0.782 ± 0.006; p > 0.05, df = 19).

In population growth models genetic differentiation was similar to moderate population density models, with a robust pattern of increasing differentiation through time in absolute barrier landscapes (FST = 0.011 ± 0.001), and semi-permeable barrier landscapes (FST = 0.003 ± < 0.001) falling between absolute barrier and neutral landscapes (FST = 0.002 ± < 0.001). At generation five there was little to no population genetic structure in all modeled landscapes, with no ostensible population structure in the neutral landscape by generation 200 (Pr(X|K) = 1, ΔK = 5). Semi-permeable (Pr(X|K) =

6, ΔK = 2) and absolute barrier landscapes (Pr(X|K) and ΔK = 2) showed strong spatial autocorrelation with genetic structure corresponding with the linear barrier by generation

200 (Fig. 2. 5).

Models with an increased number of loci (n = 80) were initiated with moderate population density (N = 8750). In the semi-permeable and absolute barrier landscapes the population decreased 8.2 – 8.3% by generation 200, similar to moderate population density models with 20 loci. The range of the mean number of alleles/locus differed from moderate population density models (A = 9.3 ± 0.1 – 11.0 ± 0.1) due to the simulation parameters used. They did conform with the simulated data in the behavior models (A = 65

9.5 – 11.0), which used the same population density and did not show significant differences by landscape scenario (p > 0.05, df = 79). There were significantly more alleles/locus at generation 200 than generation 0 in the neutral landscape (p = 2 * 10-16, df

=79). Heterozygosity remained consistent in all landscapes scenarios, with the high at generation 200 in the neutral landscape (Ho = 0.764 ± 0.006) and the low at generation

200 in the absolute barrier landscape (Ho = 0.756 ± 0.005; p > 0.05, df = 79). The neutral landscape had significantly higher heterozygosity by generation 200 when compared with generation 0 (p = 2 * 10-16, df = 79).

Models with increased number of loci showed similar patterns in genetic differentiation with other models, particularly moderate population density models. In absolute barrier landscapes genetic differentiation increased markedly through time (FST

= 0.011 ± 0.001). Semi-permeable barrier landscapes (FST = 0.003 ± <0.001) fell between absolute barrier and neutral landscapes (FST = 0.002 ± < 0.001). At generation five there was weak indication of population genetic structure in all modeled landscapes, with evidence of a cline in the neutral landscape (Pr(X|K) and ΔK = 2). The semi-permeable landscape showed strong spatial autocorrelation with a clear geographic population on either side of the linear barrier (Pr(X|K) = 4, ΔK= 2), as did the absolute barrier landscape

(Pr(X|K) and ΔK= 2; Fig. 2. 6).

Heterogeneous landscape models were initiated at moderate population density (N

= 9366). The population density decreased over 55% in all models, with the largest decline in the absolute barrier landscape (N = 4118 ± 24.3). The mean number of alleles/locus at generation 200 ranged from 19.2 ± 0.52 – 19.3 ± 0.52, and did not differ 66

significantly by landscape scenario (p > 0.05, df = 19). The number of alleles/locus decreased significantly through time (p = 1 * 10-5, df = 19). Heterozygosity decreased in all landscapes (p = 8 * 10-5, df = 19), but conformed to previous model patterns. There were no significant differences between modeled landscapes (p > 0.05, df = 19). The highest amount of heterozygosity was detected at generation 200 in the neutral landscape

(Ho = 0.739 ± 0.001) and the low in the absolute barrier landscape (Ho = 0.723 ± 0.010).

Genetic differentiation patterns corresponded with other models. In the absolute barrier landscape genetic differentiation at generation 200 (FST = 0.026 ± 0.003) was greater than predicted by the moderate population density model, but lower than at low density. Similarly, the semi-permeable barrier landscape (FST = 0.016 ± 0.002) fell between the absolute barrier and the neutral landscapes (FST = 0.006 ± 0.001). There was strong support for spatial autocorrelation by generation 200, with an inferred cline and resultant population genetic structure in the neutral landscape (Pr(X|K) = 10, ΔK = 2).

Spatial genetic structure was evident as the result of isolation in both the linear barrier landscapes (Pr(X|K) = 10, ΔK = 2), with the absolute barrier allowing for no genetic connectivity (Fig. 2. 7).

DISCUSSION

CORRIDORS IMPROVE CONNECTIVITY

Understanding how barriers contribute to declines in desert tortoise connectivity is crucial to conservation efforts (Averill-Murray et al. 2012). Barriers result in isolation of 67

populations that would otherwise not occur, ultimately increasing genetic differentiation and reducing overall genetic diversity as a result of fragmentation. Forward-in-time simulations using a broad range of population sizes and dispersal abilities found that the addition of corridors onto a fragmented landscape is predicted to improve genetic diversity for a broad range of taxa (Christie & Knowles 2015). My results indicate a decrease in the number of individuals on the landscape and some reduction in gene flow even with the addition of connectivity culverts as a result of habitat lost due to the barrier as well as increased isolation. Adding limited permeability along barriers improved gene flow but did not entirely negate the genetic effects of linear disturbances. Low levels of gene flow (i.e. one-migrant-per-generation) may suffice in preventing deleterious effects of inbreeding, but, will not adequately maintain comparable allele frequencies, or genetic connectivity, between populations (Lowe & Allendorf 2010). Additionally, negative genetic effects appear to be non-linear, with the greatest changes occurring initially following disturbance. However, my results indicate that the addition of connectivity culverts to anthropogenic barriers (road and railways) is likely to improve genetic connectivity, relative to an absolute barrier, for the desert tortoise.

The impacts of landscape change on genetic architecture is associated with a considerable lag time in detection (Anderson et al. 2010; Landguth et al. 2010). Genetic differentiation between groups on either side of an absolute barrier are likely to continue to increase with time, as these are pairwise comparisons between isolated populations.

The simulation results indicate that with any barrier (absolute or semi-permeable) structure will be evident within 200 generations of disturbance. However, genetic effects 68

may be evident in as few as five generations. Given appropriate conditions (low population density, increased number of loci) the effects of a barrier may be noticeable in fewer than five generations.

LOW POPULATION DENSITIES LOSE GENETIC CONNECTIVITY

Connectivity on the landscape was found to be heavily influenced by population density, and landscapes with lower density populations experienced greater reductions in population size and genetic diversity (heterozygosity and alleles/locus) with or without barriers, likely as the result of individuals moving across the simulated landscape without finding mates and genetic drift. Absolute barriers fragmented populations, increasing population genetic structure. When density was moderate to high, genetic diversity was largely unaffected and population size only decreased when a barrier was present. With increased density (moderate to high) barriers resulted in greater genetic differentiation and population genetic structure, but to a lesser degree than in low population density simulations. Adding permeability to barriers is predicted to improve connectivity at any population density; however, the consequences of low density may result in greater risk of genetic drift and harmful stochastic demographic processes (Dileo et al. 2013; Mateo-

Sanchez et al. 2014; Moqanaki & Cushman 2016).

MANAGEMENT RECOMMENDATIONS

Evaluating the total number of individuals and population genetic structure was useful in detecting changes in modeled scenarios, and genetic diversity was most informative at 69

low population densities. Given that tortoises were listed as threatened in 1990 due to range-wide population declines (USFWS 1994) and have continued to decline since listing (Allison & McLuckie 2018) I recommend that small changes to genetic diversity be monitored closely, as they may indicate more severe effects associated with population decline (Barr et al. 2015; Segelbacher et al. 2003; Vandergast et al. 2015;

Wood et al. 2016). Additionally, care should be taken to ensure Mojave desert tortoise populations remain connected by large intact tracts of habitat and without losing population density. In simulation studies Fahrig (2001) found a precipitous drop in the survival-habitat relationship, indicating that as a population approaches the threshold additional small losses of habitat have a large impact on the probability survival.

Therefore, the main priority for conservation is habitat protection and restoration of connectivity.

Altering the population growth rate from 0.5% to 1% annual growth did not influence genetic connectivity in this study, but future investigations could benefit from additional research on this parameter, as well as dispersal probability/migration rate estimates. A better understanding of this parameters could serve to improve future modeling efforts. While using more loci did not change model outcomes, it had the tendency for clearer detection of spatial autocorrelation and loss of heterozygosity in fewer generations (often by generation five following disturbance). Additional studies could use methods allowing for more loci or next generation sequencing techniques to improve sensitivity.

ACKNOWLEDGEMENTS

Ken Nussear and Jill Heaton provided valued support and contributed greatly to the implementation of this project. I thank Ken Nussear for working closely with me on these simulations. Discussion with Amy Vandergast, Todd Esque (USGS), and Marjorie

Matocq (UNR) contributed to the development of this research. I would also like to thank

Amy Vandergast and Anna Mitelberg (USGS) for their assistance with genotyping. I am grateful for those who contributed to the acquisition of initial data including Kristina

Drake, Felicia Chen, Ben Gottsacker, Amanda McDonald, Jordan Swart, and Sara

Murray. Funding for this project (Desert Tortoise Connectivity Modeling; 2015-UNR-

1580A) has been provided by the Bureau of Land Management through the sale of public lands as authorized by the Southern Nevada Public Land Management Act. The Clark

County Desert Conservation Program served as the funding source and point of contact for this research project. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the opinions or policies of the

U.S. Government. Mention of trade names or commercial products does not constitute their endorsement by the U.S. Government. 71

REFERENCES

Allison LJ, McLuckie AM (2018) Population Trends in Mojave Desert Tortoises (Gopherus agassizii). Herpetological Conservation and Biology 13(2):433-452

Anderson CD, Epperson BK, Fortin MJ, Holderegger R, James PMA, Rosenberg MS, Scribner KT, Spear S (2010) Considering Spatial and Temporal Scale in Landscape- Genetic Studies of Gene Flow. Molecular Ecology 19(17):3565-3575

Averill-Murray RC, Darst CR, Field KJ, Allison LJ (2012) A New Approach to Conservation of the Mojave Desert Tortoise. BioScience 62(10):893-899

Balkenhol N, Waits LP (2009) Molecular Road Ecology: Exploring the Potential of Genetics for Investigating Transportation Impacts on Wildlife. Molecular Ecology 18(20):4151-4164

Barr KR, Kus BE, Preston KL, Howell S, Perkins E, Vandergast AG (2015) Habitat Fragmentation in Coastal Southern California Disrupts Genetic Connectivity in the Cactus Wren (Campylorhynchus brunneicapillus). Molecular Ecology 24(10):2349-2363

Beier P, Majka DR, Spencer WD (2008) Forks in the Road: Choices in Procedures for Designing Wildland Linkages. Conservation Biology 22(4):836-851

Beier P, Noss RF (1998) Do Habitat Corridors Provide Connectivity? Conservation Biology 12(6):1241-1252

Bivand R, Rundel C (2019) rgeos: An Interface to Geometry Engine – Open Source (‘GEOS’), R Package Version 0.4-3. https://CRAN.R-project.org/package=rgeos

Boarman WI, Sazaki M (2006) A Highway’s Road-Effect Zone for Desert Tortoises (Gopherus agassizii). Journal of Arid Environments 65(1):94-101

Boarman WI, Sazaki M, Jennings WB (1997) The Effects of Roads, Barrier Fences, and Culverts on Desert Tortoise Populations in California, USA. In: Proceedings of Conservation, Restoration, and Management of Tortoises and Turtles – An International Conference, pp 54-58

Brooks CP (2003) A Scalar Analysis of Landscape Connectivity. Oikos 102(2):433-439

Bruford MW, Ancrenaz M, Chikhi L, Lackman-Ancrenaz I, Andau M, Ambu L, Goosens B (2010) Projecting Genetic Diversity and Population Viability for the Fragmented Orangutan Population in the Kinabatangan Floodplain, Sabah, Malaysia. Endangered Species Research 12(3):249-261

Christie MR, Knowles LL (2015) Habitat Corridors Facilitate Genetic Resilience Irrespective of Species Dispersal Abilities of Population Sizes. Evolutionary Applications 8(5):454-463

Cushman SA, McRae B, Adriaensen F, Beier P, Shirley M, Zeller K (2013) Biological Corridors and Connectivity. In: Macdonald DW, Willis KJ (ed) Key Topics in Conservation Biology 2, 1st edition, John Wiley & Sons Ltd, pp 384-404

Dileo MF, Rouse JD, Davila JA, Lougheed SC (2013) The Influence of Landscape on Gene Flow in the Eastern Massasauga Rattlesnake (Sistrurus c. catenatus): Insight from Computer Simulations. Molecular Ecology 22(17):4483-4498

Dileo MF, Wagner HH (2016) A Landscape Ecologist’s Agenda for Landscape Genetics. Current Landscape Ecology Reports 1(3):115-126

Dixo M, Metzger JP, Morgante JS, Zamudio KR (2009) Habitat Fragmentation Reduces Genetic Diversity and Connectivity Among Toad Populations in the Brazilian Atlantic Coastal Forest. Biological Conservation 142(8):1560-1569

Doak D, Kareiva P, Klepetka B (1994) Modeling Population Viability for the Desert tortoise in the Western Mojave Desert. Ecological Applications 4(3):446-460

Dutcher KE, Vandergast AG, Esque TC, Mitelberg A, Matocq MD, Heaton JS, Nussear KE (2020) Genes in Space: What Mojave Desert Tortoise Genetics Can Tell Us About Landscape Connectivity. Conservation Genetics https://doi.org/10.1007/s10592-020- 01251-z

Edwards T, Lathrop A, Ngo A, Choffe K, Murphy RW (2003) STR/Microsatellite Primers for the Desert Tortoise, Gopherus agassizii, and its Congeners. Conservation Genetics Resources 3(2):365-368

Edwards TE, Berry KH, Inman RD, Esque TC, Nussear KE, Jones CA, Culver M (2015) Testing Taxon Tenacity of Tortoises: Evidence for a Geographical Selection Gradient at a Secondary Contact Zone. Ecology and Evolution 5(10):2095-2114

Epperson BK, McRae BH, Scribner K, Cushman SA, Rosenberg MS, Fortin MJ, James PMA, Murphy M, Manel S, Legendre P, Dale MRT (2010) Utility of Computer Simulations in Landscape Genetics. Molecular Ecology 19(17):3549-3564

Estoup A, Angers B (1998) Microsatellites and Minisatellites for Molecular Ecology: Theoretical and Empirical Considerations. In: Carvalho G (ed) Advances in Molecular Ecology, IOS Press Amsterdam, pp 55-86

Evanno G, Regnaut S, Goudet J (2005) Detecting the Number of Clusters of Individuals using the Software STRUCTURE: A Simulation Study. Molecular Ecology 14(8):2611- 2620

Fahrig L (2001) How Much Habitat is Enough? Biological Conservation 100(1):65-74

Francis RM (2017) PopHelper: An R Package and Web App to Analyze and Visualize Population Structure. Molecular Ecology Resources 17(1):27-32

Gregory AJ, Beier P (2014) Response Variables for Evaluation of the Effectiveness of Conservation Corridors. Conservation Biology 28(3):689-695

Goudet J (2005) Hierfstat, a Package for R to Compute and test Variance Components and F-Statistics. Molecular Ecology Notes 5:184-186

Haddad NM, Bowne DR, Cunningham A, Danielson BJ, Levey DJ, Sargent S, Spira T (2003) Corridor Use by Diverse Taxa. Ecology 84(3):609-615 75

Hagerty BE, Peacock MM, Kirchoff VS, Tracy CR (2008) Polymorphic Microsatellite Markers for the Mojave Desert Tortoise, Gopherus agassizii. Molecular Ecology Resources 8(5):1149-1151

Henein K, Merriam G (1990) The Elements of Connectivity where Corridor Quality is Variable. Landscape Ecology 4(2-3):157-170

Hijmans RJ (2019) raster: Geographic Data Analysis and Modeling, R Package Version 2.9-5. https://CRAN.R-project.org/package=raster

Holderegger R, Kamm U, Gugerli F (2006) Adaptive vs. Neutral Genetic Diversity: Implications for Landscape Genetics. Landscape Ecology 21(6):797-807

Holderegger R, Di Giulio M (2010) The Genetic Effects of Roads: A Review of Empirical Evidence. Basic and Applied Ecology 11(6):522-531

Holderegger R, Wagner HH (2008) Landscape Genetics. BioScience 58(3):199-207

Jombart T (2008) adegenet: A R Package for the Multivariate Analysis of Genetic Markers. Bioinformatics 24(11):1403-1405

Jombart T, Dufour AB, Pontier D (2008) Revealing Cryptic Spatial Patterns in Genetic Variability by a New Multivariate Method. Heredity 101(1):92-103

Kahle D, Wickman H (2013) ggmap: Spatial Visualization with ggplot2. The R Journal 5(1):144-161

Landguth EL, Cushman SA, Schwartz MK, McKelvey KS, Murphy M, Luikart G (2010) Quantifying the Lag Time to Detect Barriers in Landscape Genetics. Molecular Ecology 19(19):4179-4191

Lowe WH, Allendorf FW (2010) What can Genetics tell us about Population Connectivity? Molecular Ecology 19(15):3038-3051

Manel S, Schwartz MK, Luikart G, Taberlet P (2003) Landscape Genetics: Combining Landscape Ecology and Population Genetics. Trends in Ecology and Evolution 18(4):189-197

Mateo-Sanchez MC, Cushman SA, Saura S (2014) Connecting Endangered Brown Bear Subpopulations in the Cantabrian Range (North-Western Spain). Animal Conservation 17(5):430-440

McCoy ED, Aguirre G, Kazmaier RT, Tracy CR (2014) Demography of North American Tortoises. In: Rostal DC, McCoy ED, Mushinsky HR (ed) Biology & Conservation of North American Tortoises, John Hopkins University Press, pp 143-148

Medica PA, Nussear KE, Esque TC, Saethre MB (2012) Long-Term Growth of Desert Tortoises (Gopherus agassizii) in a Southern Nevada Population. Journal of Herpetology 46(2):213-220

Meirmans PG (2015) Seven Common Mistakes in Population Genetics and How to Avoid Them. Molecular Ecology 24(13):3223-3231

Moqanaki EM, Cushman SA (2016) All Roads Lead to Iran: Predicting Landscape Connectivity of the Last Stronghold for the Critically Endangered Asiatic Cheetah. Animal Conservation 20(1):29-41 77

Nafus MG, Tuberville TD, Buhlmann KA, Todd BD (2013) Relative Abundance and Demographic Structure of Agassiz’s Desert Tortoise (Gopherus agassizii) along Roads of Varying Size and Traffic Volume. Biological Conservation 162:100-106

Nei M (1973) Analysis of Gene Diversity in Subdivided Populations. In: Proceedings of the National Academy of Sciences 70(12):3321-3323

Noss RF (1987) Corridors in Real Landscapes: A Reply to Simberloff and Cox. Conservation Biology 1(2):159-164

Nussear KE, Tracy CR, Medica PA, Wilson DS, Marlow RW, Corn PS (2012) Translocation as a Conservation Tool for Agassiz’s Desert Tortoises: Survivorship, Reproduction, and Movements. The Journal of Wildlife Management 76(7):1341-1353

Peaden JM, Nowakowski AJ, Tuberville TD, Buhlmann KA, Todd BD (2017) Effects of Roads and Roadside Fencing on Movements, Space Use, and Carapace Temperatures on a Threatened Tortoise. Biological Conservation 214:13-22

Pritchard J K, Stephens M, Donnelly P (2000) Inference of Population Structure using Multilocus Genotype Data. Genetics 155(2):945-959

Pritchard JK, Wen X, Falush D (2010) Documentation for STRUCTURE Software: Version 2.3. University of Chicago, Chicago, pp 1-39 78

R Core Team (2019) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/

Rautsaw RM, Martin SA, Vincent BA, Lanctot K, Bolt MR, Seigel RA, Parkinson CL (2018) Stopped Dead in Their Tracks: The Impact of Railways on Gopher Tortoise (Gopherus Polyphemus) Movement and Behavior. Copeia 106(1):135-143

Rebaudo F (2014) SIMADAPT (Version 1.8.0). CoMSES Computational Model Library https://www.comses.net/codebases/3137/releases/1.8.0/

Rebaudo F, Cost J, Almeida CE, Silvain JF, Harry M, Dangles O (2014) Simulating Population Genetics of Pathogen Vectors in Changing Landscapes: Guidelines and Application with Triatoma brasiliensis. PLOS Neglected Tropical Diseases 8(8):1-8

Rebaudo F, Le Rouzic A, Dupas S, Silvain JF, Harry M, dangles O (2013) SIMADAPT: An Individual-Based Genetic Model for Simulating Landscape Management Impacts on Populations. Methods in Ecology and Evolution 4(6):595-600

Segelbacher G, Hoglund J, Storch I (2003) From Connectivity to Isolation: Genetic Consequences of Population Fragmentation in Capercaillie across Europe. Molecular Ecology 12(7):1773-1780

Slatkin M (1985) Gene Flow in Natural Populations. Annual Review of Ecology and Systematics 16(1):393-430

Sork VL, Waits L (2010) Contributions of Landscape Genetics – Approaches, Insights, and Future Potential. Molecular Ecology 19(17):3489-95

Storfer A, Murphy MA, Evans JS, Goldberg CS, Robinson S, Spear SF, Dezzani R, Delmelle E, Vierling L, Waits LP (2007) Putting the ‘Landscape’ in Landscape Genetics. Heredity 98(3):128-142

Schwartz MK, McKelvey KS (2008) Why Sampling Scheme Matters: The Effect of Sampling Scheme on Landscape Genetic Results. Conservation Genetics doi:10.1007/s10592-008-9622-1

Schwartz TS, Osentoski M, Lamb T, Karl SA (2003) Microsatellite Loci for the North American Tortoises (Genus Gopherus) and their Applicability to Other Turtle Species. Molecular Ecology Notes 3(2):283-286

Taylor PD, Fahrig L, Henein K, Merriam G (1993) Connectivity is a Vital Element of Landscape Structure. Oikos 68(3):571-573

Tracy CR, Averill-Murray R, Boarman WI, Delehanty D, Heaton J, McCoy E, Morafka DJ, Nussear KE, Hagerty B, Medica P (2004) Desert Tortoise Recovery Plan Assessment. US Fish & Wildlife Service, Desert Tortoise Recovery Plan Assessment Committee

Trombulak SC, Frissell CA (2000) Review of Ecological Effects of Roads on Terrestrial and Aquatic Communities. Conservation Biology 14(1):18-30

Turner FB, Hayden BL, Burge BL, Roberson JB (1986) Egg Production by the Desert Tortoise (Gopherus agassizii) in California. Herpetologica 42:93-104 80

USFWS (US Fish & Wildlife Service) (1994) Desert Tortoise (Mojave Population) Recovery Plan. US Fish & Wildlife Service, Portland, OR

USFWS (US Fish & Wildlife Service) (2011) Revised Recovery Plan for the Mojave Population of the Desert Tortoise (Gopherus agassizii). Pacific Southwest Region, Sacramento, CA

Vandergast AG, Wood DA, Thompson AR, Fisher M, Barrows CW, Grant TJ (2015) Drifting to Oblivion: Rapid Genetic Differentiation in an Endangered Lizard Following Habitat Fragmentation and Drought. Diversity and Distributions 22(3):1-14 vonSekendorff Hoff K, Marlow RW (2002) Impacts of Vehicle Road Traffic on Deseret Tortoise Populations with Consideration of Tortoise Habitat in Southern Nevada. Chelonian Conservation and Biology 4(2):449-456

Wilensky U (1999) NETLOGO. Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston IL http://ccl.northwestern.edu/netlogo/

Wood DA, Bui TVD, Overton CT, Vandergast AG, Casazza ML, Hull JM, Takekawa JY (2016) A Century of Landscape Disturbance and Urbanization of the San Francisco Bay Region Affects Present-Day Genetic Diversity of the California Ridgeway Rail (Rallus obsoletus obsoletus). Conservation Genetics 18(1):131-136

Zeller K, McGarigal K, Whiteley A (2012) Estimating Landscape Resistance to Movement: A Review. Landscape Ecology 27(6):777-97 81

TABLES AND FIGURES

Table 2. 1. Simulation parameters. Surface: (S) simple resistance surface with no resistance outside barriers; (H) heterogeneous resistance surface (0-0.6) outside barriers. Cells: number of grid cells per simulation. Density: carrying capacity in each 1 km2 grid cell; (Variable) varies with resistance surface (range 1 – 24). N: number of tortoises used to seed simulations. r: population growth rate. Data: (IG) initial genotypes from a panmictic population; (GP) genetic parameters from initial genotypes; (SG) simulated genotypes. Loci: number of loci. Reps: number of repetitions per simulation. No: number of models run.

Model Surface Cells Density N r Data Loci Reps No

Genotypes S 625 24 170 0.48 IG 20 1 1 Behavior S 625 14 8750 0.48 GP 20 1 3 Limitations S 1050 14 14700 0.48 GP 20 1 1 Density Low S 625 3 1875 0.48 SG 20 30 3

Mod S 625 14 8750 0.48 SG 20 30 3

High S 625 24 15000 0.48 SG 20 30 3

Growth S 625 14 8750 0.24 SG 20 30 3 Loci S 625 14 8750 0.48 GP 80 30 3 Landscape H 625 Variable 9366 0.48 SG 20 30 3

Table 2. 2. Carrying capacities determined by habitat suitability values (HSV) from a desert tortoise habitat suitability model (Nussear et al. 2009). Resistance values calculated as inverse of the HSV by each grid cell.

HSV Resistance Carrying Capacity

0 1.00 1

0.01 – 0.10 0.90 – 0.99 1

0.11 – 0.20 0.80 – 0.89 1

0.21 – 0.30 0.70 – 0.79 3

0.31 – 0.40 0.60 – 0.69 6

0.41 – 0.50 0.50 – 0.59 9

0.51 – 0.60 0.40 – 0.49 12

0.61 – 0.70 0.30 – 0.39 15

0.71 – 0.80 0.20 – 0.29 18

0.81 – 0.90 0.10 – 0.19 21

0.91 – 1.00 0 – 0.09 24

Table 2. 3. Genetic statistics for genotypes used in simulations: (n) sample size; (A) alleles/locus; (Ho) mean observed heterozygosity; (He) mean expected heterozygosity;

(FST) genetic differentiation.

Genotypes n A Ho ± SD He ± SD FST

Original 170 14.4 0.80 ± 0.13 0.81 ± 0.12 0.003

Simulated (subsample) 750 23.2 0.79 ± 0.11 0.82 ± 0.11 0.001

Simulated (all) 14572 27.7 0.79 ± 0.11 0.82 ± 0.11 < 0.001

Fig. 2. 1. Simulation model landscapes. A) Left to right: neutral landscape with no resistance (0), semi-permeable barrier (0.7 and 1), absolute vertical barrier (1), absolute horizontal barrier (1). Neutral, absolute vertical barrier, and absolute horizontal barrier landscapes used in behavior models; neutral, semi-permeable, and absolute vertical barrier used in population density, population growth, and increased number of loci models. B) Left to right: neutral landscape with heterogeneous resistance values (0-0.6), semi-permeable barrier (0.7 and 1), absolute barrier (1). Landscapes used in heterogeneous landscape models. 85

Fig. 2. 2. Map of tortoise locations from a continuous population in Ivanpah Valley, along the California/Nevada border, used as the initial genetic data in forward-in-time simulations (n = 170). 86

Fig. 2. 3. Population genetic structure for genotypes used to seed simulation models. Left to right: A) sPCA results from the original genetic data (n = 170) randomized with no identifiable genetic pattern; a subsample of simulated genotypes (n = 750) showing a genetic cline; all simulated genotypes (N = 14572) showing a genetic cline. B)

STRUCTURE results from the original genetic data (n = 170) randomized (Pr(X|K) = 2, ΔK = 5); a subsample of simulated genotypes (n = 750; Pr(X|K) = 1, ΔK = 3). 87

Fig. 2. 4. Population density model results using (upper) low density of 3 animals/km2 animals/km2, (mid) moderate at 14 animals/km2, (lower) high at 20, each 625 km2. Results reported for three modeled landscapes: neutral with no resistance, semi- permeable barrier, absolute barrier. Left to right: times-series of A) number of individuals

(N), observed heterozygosity (Ho), genetic differentiation (FST); B) population genetic pattern results from sPCA. C) STRUCTURE results. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 89

Fig. 2. 5. Population growth model results using moderate density (14 animals/km2) and r = 0.24 (compare with results of moderate density (14 animals/km2) and r = 0.48 in Fig. 2. 4). Results reported for three modeled landscapes: neutral with no resistance, semi- permeable barrier, absolute barrier. Left to right: times-series of A) number of individuals

(N), observed heterozygosity (Ho), genetic differentiation (FST); B) population genetic pattern results from sPCA; C) STRUCTURE results. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 90

Fig. 2. 6. Increased number of loci model results using moderate density (14 animals/km2) and 80 loci (compare with results of moderate density (14 animals/km2) using 20 loci in Fig. 2. 4). Results reported for three modeled landscapes: neutral with no resistance, semi-permeable barrier, absolute barrier. Left to right: times-series of A) number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) population genetic pattern results from sPCA; C) STRUCTURE results. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 91

Fig. 2. 7. Heterogeneous landscape model results reported for three modeled landscapes: neutral with no resistance, semi-permeable barrier, absolute barrier. Landscapes were modeled at 625 km2 with number of animals/km2 dependent on cell resistance. Left to right: times-series of A) number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) population genetic pattern results from sPCA; C) STRUCTURE results. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 92

CHAPTER 3. GENES THROUGH TIME: HOW PREDICTIONS OF MOJAVE

DESERT TORTOISE GENETICS CAN INFORM CONNECTIVITY TODAY

ABSTRACT

Fragmentation and habitat loss reduce population sizes, impede connectivity, and are primary threats to biodiversity. Urbanization and large-scale solar development in

Southern Nevada mainly occur in Mojave desert tortoise (Gopherus agassizii) habitat, a species that has shown rapid population declines associated with habitat loss and degradation. Little work has been done to examine how habitat loss and fragmentation impede tortoise connectivity. This study used individually-based spatially explicit forward-in-time simulations to predict genetic connectivity in no disturbance and disturbance scenarios. I modeled 17 areas in Clark County, Nevada that were 525 – 625 km2 in area, at a 1 km2 resolution. These models used resistance surfaces that included potential disturbance factors that could impede the connectivity of desert tortoise populations. I examined population size, genetic diversity (alleles/locus, heterozygosity), and genetic structure (FST, sPCA, STRUCTURE) through time in each landscape scenario. I used a connectivity success index, based on genetic differentiation, to forecast the success or failure of gene flow over time, and translated this metric into structural landscape patterns. As anthropogenic disturbance increased, so did demographic and genetic effects. Habitat degradation resulted in predicted population declines, with the most pronounced losses concomitant with increased disturbance. Genetic diversity was 93

increasingly lost as disturbance intensified. Because connectivity success was found to be landscape dependent, outcomes for maintaining genetic connectivity were variable, but predicted gene flow was always reduced with disturbance. Disturbance landscapes with high levels of genetic connectivity tended towards low levels of fragmentation and landscape complexity, while the amount and dominance (based on largest patch size) of suitable habitat remained high. My results indicate that adequately protecting existing intact tracks of tortoise habitat and ensuring sufficient connectivity will benefit species recovery.

INTRODUCTION

Nevada ranks first in the United States in human population growth rate, with increases of 12.4% since 2010 (USCB 2018). The bulk of this growth is within Clark County, which has experienced increases of greater than 40% since 2000 (Pendleton et al. 2013), and Las Vegas and Henderson are among the top 15 fastest growing cities in the nation

(USCB 2018). Using aerial imagery from 2017 to detect the total land amount with evidence of disturbance in Clark County, it was estimated that 1,100 km2 within the county has been developed (Clark County 2017). Additionally, large-scale solar facilities have been developed, or approved for development, on 60 km2 within the county, chiefly on federal lands (BLM 2018a; 2018b; 2018c). Urbanization does not occur without significant environmental impact, and in southern Nevada, land use related to human 94

population growth, and energy development primarily occurs in lower elevation ecosystems like Mojave Desert scrub (Clark County 2017; Pendleton et al. 2013).

Habitat disturbance has been shown to reduce population sizes, and is a primary threat to biodiversity (Fahrig 2003; Haddad et al. 2015). Disturbance poses risks to the long-term persistence of species like the Mojave desert tortoise (Gopherus agassizii) which has shown rapid population declines associated with habitat loss and degradation due to urbanization (Allison & McLuckie 2018; Doak et al. 1994; Corn 1994; Tracy et al.

2004; USFWS 2011), resulting in federal listing as a threatened species in 1990 (USFWS

1994). Threats to tortoise populations continue to intensify as land is converted for human uses, elevating the need to protect conservation areas and corridors between them that facilitate connectivity (Averill-Murray et al. 2012). Established tortoise conservation areas are limited by current land ownership and land use designations. Habitat loss and fragmentation can impede connectivity, disrupting species interactions, altering landscape use, and reducing rescue effects (Ewers & Didham 2006; Haddad et al. 2017; Hansen et al. 2007; Hand et al. 2014). Range-wide 16,282 km2 of habitat lie outside conservation areas and are subject to growing development pressures (Carter et al. in review), making the need to maintain connective habitat for tortoise populations through the Mojave

Desert critical (Averill-Murray et al. 2013).

The conservation value of a habitat corridor lies in its ability to provide functional connectivity (Beier & Noss 1998). Appropriate corridors are determined by habitat selection and movement (Chetkiewicz et al. 2006), with features of primary importance determined by multiple factors (e.g. length, width, topography, vegetation, adjacency to 95

human activities, and the habitat needs of the species of interest), and not necessarily based on a simple measurement of area (Beier & Loe 1992; Noss 1987). For taxa with limited dispersal, such as tortoises, corridors are likely not used for swift movement events or migrations; rather individuals may need days to generations to achieve connectivity (Beier & Loe 1992). Beier et al. (2008) suggest that corridor dwelling species may require corridor widths considerably larger than home-range width to provide most, if not all, ecological needs.

The Mojave desert tortoise is commonly associated with desert scrub, but is known to occupy and move through heterogeneous habitat (Morafka & Berry 2002), and has been recorded in rugged terrain (O’Connor et al. 1994; Dutcher et al. 2020). For desert tortoises a historically well connected landscape with few barriers to movement has resulted in a range-wide pattern of isolation-by-distance (IBD) with gene flow and little genetic differentiation (Britten et al. 1997; Murphy et al. 2007; Hagerty & Tracy

2010; Hagerty et al. 2011; Sanchez-Ramirez et al. 2018; Shaffer et al. 2015). Landscape genetics are sensitive to temporal scale, often revealing gene flow on a historic landscape even if it no longer occurs (Waples & Gaggiotti 2006). While the effects of fragmentation can be strong and persistent, their timescale is uncertain (Haddad et al. 2017). Landscape genetics studies find the timescale to detect the effects of habitat disturbance and the creation of barriers on the landscape can be decades to millennia (Gauffre et al. 2015;

Gonzales et al. 2010; Leblois et al. 2004; McRae et al. 2005; Murphy et al. 2008; Row et al. 2011; Segelbacher et al. 2003). Because desert tortoises are estimated to reach reproductive age at 17 years, on average (USFWS 2011), directly evaluating the 96

consequences of recent or planned habitat loss and fragmentation on genetic connectivity requires investigations of potential future patterns of population genetics.

Spatially explicit forward-in-time simulations can merge projections of changing conditions with landscape genetics to predict population persistence and functional connectivity in the future (Epperson et al. 2010; Creech et al. 2017; Rebaudo et al. 2014;

Thatte et al. 2018). However, spatially correlated natural and anthropogenic features complicate inference, making interpretation difficult without comparative study design

(Beier & Noss 1998). For example, anthropogenic barriers, like roads, can reduce gene flow, but the effects may be confounded by long-standing natural features (Dileo et al.

2013; Vandergast et al. 2007). Evaluation of connectivity with and without habitat disturbance can disentangle these effects (Beier & Noss 1998), and forward-in-time simulations can be used to model multiple landscape scenarios.

This study used forward-in-time simulation modeling to predict tortoise genetic connectivity in Clark County, Nevada using three landscape scenarios: (1) no habitat disturbance, (2) current levels of habitat disturbance, and (3) future projections of habitat disturbance given by development scenarios currently under consideration. In quantifying the genetic effects of current and planned habitat disturbance on populations through time, I sought to uncover what constitutes connectivity success by investigating the shared characteristics of landscapes projected to maintain genetic connectivity for

Mojave desert tortoises into the future.

MATERIALS AND METHODS 97

STUDY LANDSCAPE RESISTANCE SURFACES

The study landscape focused on Clark County, Nevada using resistance surfaces. The neutral representation without anthropogenic disturbance was calculated as the inverse of an existing desert tortoise habitat suitability model (illustrating range-wide habitat potential and key areas for connectivity in the absence of anthropogenic disturbance,

Nussear et al. 2009). Resistance surface landscapes incorporating current levels of habitat disturbance future projections of disturbance based on a 50 year forecast (Fig. 3. 1; forecast data provided by Clark County) were also created. Landscapes were mapped using R 3.5.3 (R Core Team 2019) packages ggmap v.3.0.0.901, raster v.2.9-5, and rgeos v.0.4-3 (Kahle & Wickman 2013; Hijmans 2019; Bivand & Rundel 2019). Within the study landscape I modeled areas 525 – 625 km2 with a 1 km2 resolution at 17 locations, using the three landscape resistance scenarios above, plus a hypothetical neutral landscape with an absolute barrier to dispersal based on existing landscape features/projected disturbance (see Fig. 3. 2 for each landscape). Any location bisected by a political boundary was allowed to include areas outside Clark County.

To account for habitat disturbance, I used conversion factors as values of possible degradation to adjust habitat suitability values prior to taking the inverse for resistance.

Scale factors were adapted for desert tortoises based on those used by Inman et al. (2013) as:

ℎ푎푏𝑖푡푎푡 푠푢𝑖푡푎푏𝑖푙𝑖푡푦 − (ℎ푎푏𝑖푡푎푡 푠푢𝑖푡푎푏𝑖푙𝑖푡푦 ∗ 푐표푛푣푒푟푠𝑖표푛 푓푎푐푡표푟) 98

Because habitat disturbance associated with urbanization, human access, and off highway vehicles are considered range-wide threats to desert tortoises (Darst et al. 2013) I categorized habitat disturbance as urban/cleared land, solar, railway, major roads, minor roads (e.g. unpaved), and utility corridor right-of ways (ROWs). Conversion factors were applied to each category to simulate reduced habitat suitability in disturbed areas (Table

3. 1). I assumed that urban/cleared land and fenced solar facilities represent a complete loss of habitat, as tortoises have been extirpated from large areas of their range in and near cities and towns (USFWS 1994) and large-scale solar development typically includes complete removal of vegetation, grading, and fencing (Lovich & Ennen 2011).

Formidable linear features (major roads and railways) fragment habitat and tortoise populations are depressed several hundreds of meters from roadways, likely due to extended periods of elevated mortality (Boarman & Sazaki 2006; vonSekendorff Hoff &

Marlow 2002); therefore, I expected major roads and railroads to represent substantial, but not complete, loss of habitat (Rautsaw et al. 2018). I applied a relatively low maximum conversion factor to minor roads based on length, so cells with higher road density were associated with higher penalties. I assumed a fairly low conversion factor for ROWs. Both minor roads and ROWs are generally unfenced with greater abundance of tortoise sign than well-traveled paved roads, but still with detectable impacts (Nafus et al. 2013).

MODELING FRAMEWORK AND PARAMETERS

Landscape genetic processes were simulated forward-in-time using SIMADAPT v.1.8.0

(Rebaudo 2014), an individually-based model of mating and dispersal. SIMADAPT is spatially explicit, using resistance surfaces, where each cell in the raster is assigned a carrying capacity. I simulated gene flow for 200 non-overlapping generations, sampling individuals every five generations. I performed 30 replicate runs with each landscape resistance surface, to evaluate average outcomes.

I used a previously collected genetic dataset from tortoises within the Ivanpah

Valley (Dutcher et al. 2020), amplified at 20 variable microsatellite loci previously developed for tortoises (Edwards et al. 2003; Hagerty et al. 2008; Schwartz et al. 2003).

The samples were randomized to create a group devoid of population genetic structure.

The genotypes were then simulated forward-in-time using SIMADAPT on a landscape surface without resistance with a burn-in period of 100 generations. The resultant dataset was subsampled and used as input for landscape simulation models.

Landscape simulations were initiated with a mutation rate of 0.0005 per locus per generation (Dileo et al. 2013; Estoup & Angers 1998; Landguth et al. 2010). Population growth rate was set at r = 0.48 (derived from 1% annual population growth; USFWS

1994). Individuals could move as much as 14 km in one generation (up to 10 grid cells).

Probability of dispersal from each 1 km2 raster cell was estimated at 50% (Hromada pers comm.). The carrying capacity for each cell in the landscape was determined by the relationship between habitat suitability and population density (Nussear et al. in prep) and ranged from 0 tortoises in cells with a resistance value of 1, to 24 animals in cells with no 100

resistance. Simulations were run to generate genotype files and model 17 Clark County,

Nevada locations using four landscape scenarios, resulting in 68 simulation models.

POPULATION SIZE, GENETIC DIVERSITY, AND GENETIC STRUCTURE

To subset the simulated dataset for analyses I randomly sampled without replacement (n

= 750). Simulated data were examined using average outcomes. I evaluated population size through time in neutral landscape models with disturbance models, both current disturbance and future projections of disturbance. Genetic diversity was calculated using the number of alleles/locus (A) and observed heterozygosity (Ho) in the R package adegenet v.2.1.1 (Jombart 2008).

Population genetic structure was determined by pairwise genetic differentiation

(FST, Nei 1973) in the R package hierfstat v.0.04-22 (Goudet 2005), spatial principal components analysis (sPCA, Jombart et al. 2008), and a Bayesian clustering analysis

(STRUCTURE v.2.3.4, Pritchard et al. 2000). Genotype files best representing the mean FST were used for STRUCTURE analyses. I used the admixture model, with correlated allele frequencies, and location as a prior. I estimated the probability of K population clusters =

1-10 using ten replicate runs of 1,000,000 Markov Chain Monte-Carlo iterations (burn-in

= 500,000). Because the mean log probability of the data (Pr(X|K) in Pritchard et al.

2000) may overestimate genetic clusters with IBD I also calculated the second order rate of change (ΔK in Evanno et al. 2005). Results were visualized using PopHelper v.2.2.9 in

R (Francis 2017). Finally, I used sPCA to evaluate genetic patterns that may be difficult to detect in the presence of IBD using randomized Monte-Carlo permutations (999). 101

QUANTIFYING CONNECTIVITY SUCCESS

I used a connectivity success index to forecast how well disturbed landscapes maintain gene flow over time. The connectivity success index was based on average pairwise FST outcomes from forward-in-time simulations. Pairwise genetic differentiation values from disturbed landscape locations was compared with pairwise differentiation in the neutral

(connected) landscape and in the hypothetical absolute barrier (isolated) landscape. The connectivity success index for current and disturbed landscapes was adapted from

Gregory & Beier (2014) and calculated using FST for:

(푑𝑖푠푡푢푟푏푎푛푐푒 − 𝑖푠표푙푎푡푒푑) / (푐표푛푛푒푐푡푒푑 − 𝑖푠표푙푎푡푒푑)

Values near 1 indicate gene flow comparable to the neutral landscape, while values below

0 indicate failure to maintain genetic connectivity. Because disturbed landscapes may show increased habitat loss, degradation, and fragmentation relative to the isolated

(hypothetical barrier) landscape, values below 0 may also be possible. Because of the lag time for genetic divergence, it is possible to falsely assume gene flow is retained if measured too soon. Therefore, I report values at generation 200 to ensure capture of possible genetic differentiation.

Disturbed landscapes were ranked by ability to maintain genetic connectivity based on their connectivity success index value into one of three categories, that were determined ad-hoc by dividing index values roughly into thirds: high = 1-0.70, intermediate = 0.69-0.35, low/no genetic connectivity = 0.34-negative index values. To assess the influence of landscape spatial patterns on genetic connectivity I quantified 102

landscape metrics in neutral landscapes, with current disturbance, and projections of future disturbance. Habitat suitability values (Nussear et al. 2009) were used to designate binary landscape classes: suitable habitat = 1-0.3; unsuitable habitat = 0.2-0. All metrics assumed queens case for cell connectivity (8 directions) and landscape boundaries were not included in edge counts. I characterized categorical landscape patterns using landscapemetrics v.1.2.1 in R (Hesselbarth et al. 2019). Translation of functional connectivity to structural metrics often misses crucial aspects of landscape pattern

(Kupfer 2012). To increase ecological relevance, I used a combination of metrics that are considered strong descriptors of landscape pattern and consistent in interpretation

(Cushman et al. 2008). I evaluated if/how landscape fragmentation (number of patches by class), configuration (edge density), composition (percentage area and percentage core area by class), and dominance (largest patch index) differed by connectivity success index rank.

I compared metrics of landscape disturbance (number of suitable habitat patches, largest suitable habitat patch index, percentage suitable habitat area) with population size and genetic statistics. I used Akaike’s information criterion (AIC) to rank the strength of relationships. Differences between disturbance and neutral landscape values were used to determine losses in individuals, alleles/locus, and heterozygosity, and increases in genetic differentiation.

RESULTS

103

I modeled 17 landscape locations in Clark County, Nevada using a neutral landscape scenario, current disturbance levels, and future projections of disturbance. I examined each landscape scenario for total number of individuals, genetic diversity, and genetic differentiation in time-series. The number of individuals was highest in neutral landscapes, and decreased as habitat was lost to disturbance. Genetic diversity statistics

(A, Ho) exhibited a similar pattern. Genetic differentiation and population genetic structure was lowest in neutral landscapes, with a tendency to increase with disturbance.

Significant differences (p-values) are based on two-sided t-tests between disturbance scenarios and the neutral landscape, and reported with 19 degrees of freedom. I also examined population genetic structure with genotype files best representing the mean FST value at generation 200 for STRUCTURE and sPCA (Table 3. 2; see specific sections below for each landscape result).

BOULDER CITY CONSERVATION EASEMENT NORTH

Current disturbance simulations predicted a 22.7% loss in the number of individuals, decreasing by 40.9% with future disturbance. Based on the 17 locations modeled. this location ranked 8TH (current disturbance) and 5TH (future disturbance) in loss of population size. Compared with the neutral landscape this location lost an average of

4.9% (p = 0.002) and 9.0% (p = 1*10-12) alleles/locus in current and future disturbance

-7 scenarios. This location was predicted to lose 2.8% (p = 0.023) and 7.8% (p = 6*10 ) Ho in current and future disturbance scenarios. Genetic differentiation in the neutral landscape was stable, and the current disturbance scenario showed indications of reaching 104

stability. Genetic structure analyses supported one cluster in the neutral landscape with evidence of a cline. Spatial autocorrelation was predicted to increase with disturbance, with gene flow largely absent in future projections (Fig. 3. 2. 1).

COYOTE SPRINGS

Current disturbance simulations predicted a 13.6% loss in population size at Coyote

Springs, decreasing 17.1% with future disturbance. Based on the 17 locations modeled, this location ranked 12TH in loss of population size in both current and future disturbance scenarios. Compared with the neutral landscape this location lost 2.8% (p >0.05) and

3.5% (p >0.05) alleles/locus in current and future disturbance scenarios. This location lost

2.0% (p = 0.025) and 2.5% (p = 0.025) Ho in current and future disturbance scenarios, relative to the neutral landscape. Spatial autocorrelation was present in the neutral landscape, with evidence for two clusters with admixture. Spatial autocorrelation increased in the current disturbance scenario and was further amplified with future projections. In both disturbance scenarios there was a reduction in gene flow compared with the neutral landscape (Fig. 3. 2. 2).

DRY LAKE

A loss of 24.2% in population size was predicted in the current disturbance simulation, and 30.2% with future disturbance. Dry Lake ranked 5TH (current disturbance) and 8TH

(future disturbance) in terms of loss in population size, relative to the 17 locations modeled. The mean number of alleles/locus decreased 4.8% (current disturbance, p 105

>0.05) and 6.0% (future disturbance, p = 0.008) compared with the neutral landscape.

-4 Dry Lake lost 3.1% (p = 0.002) and 4.1% (p = 1*10 ) Ho in current and future disturbance scenarios, relative to the neutral landscape. Only the neutral landscape showed relatively stable values for genetic differentiation. STRUCTURE and sPCA analyses supported a single cluster with IBD. Strong spatial autocorrelation was apparent in current and future disturbance scenarios with support for population genetic structure related to landscape barriers. Current disturbance is predicted to result in two clusters with admixture. The future disturbance scenario lost connectivity (Fig. 3. 2. 3).

ELDORADO VALLEY

Population size losses of 7.8% were predicted with current disturbance, and 11.7% with future disturbance simulations. This location ranked 15TH (current disturbance) and 16TH

(future disturbance) in terms of loss of number of individuals, relative to the 17 locations modeled. Alleles/locus were predicted to be lost by 1.4% (current disturbance, p >0.05) and 3.4% (future disturbance, p >0.05) compared with the neutral landscape. In current and future disturbance scenarios this location lost 0.7% (p >0.05) and 2.2% (p = 8*10-6)

Ho through time, relative to the neutral landscape. Genetic differentiation in the neutral landscape was stable through time, and the current and future disturbance scenarios reached relatively steady levels. Spatial autocorrelation was present in the neutral landscape, which displayed weak population genetic structure with admixture as the result of landscape features. Spatial autocorrelation remained in current and future 106

disturbance scenarios, with increased population genetic structure and reduced admixture

(Fig. 3. 2. 4).

INDIAN SPRINGS

Population size losses of 22.2% were predicted with current disturbance, and 23.7% with future disturbance. In terms of population size loss through time this location ranked 9TH

(current disturbance) and 10TH (future disturbance), relative to the 17 locations modeled.

Compared with the neutral landscape this location lost 4.4% (p >0.05) and 5.8% (p

>0.05) alleles/locus in current and future disturbance scenarios. Relative to the neutral

-4 landscape this location lost 4.0% (p = 0.011) and 4.3% (p = 4*10 ) Ho in current and future disturbance scenarios. Genetic differentiation in the neutral landscape was stable.

Spatial autocorrelation was present in all landscape scenarios. Population genetic structure was weakly present in the neutral landscape, and amplified by anthropogenic disturbance in both current and future scenarios. All landscapes provided support for admixture; however, it was reduced by disturbance (Fig. 3. 2. 5).

IVANPAH VALLEY

A population size loss of 23.7% was predicted in current disturbance simulations. Future disturbance predicted a loss of 29.6%. Based on the 17 locations modeled, this location ranked 6TH (current disturbance) and 9TH (future disturbance) in terms of loss in number of individuals. This location lost 6.0% (current disturbance, p >0.05) and 6.6% (future disturbance, p >0.05) alleles/locus compared with the neutral landscape. Relative to the 107

neutral landscape, this location lost 4.0% (p >0.05) and 4.6% (p = 0.017) in current and future disturbance scenarios. Only the neutral landscape showed indications of steady genetic differentiation through time. The neutral landscape exhibited spatial autocorrelation, with genetic structure supporting two clusters. The disturbance scenarios are expected to result in an almost complete loss of connectivity, with the current disturbance landscape suggesting an increase in clustering and the future disturbance landscape resulting in two isolated clusters (Fig. 3. 2. 6).

JEAN/ROACH

Population size losses of 21.2% and 39.0% were predicted with current and future disturbance simulations, ranking this location 10TH (current disturbance) and 6TH (future disturbance), based on the 17 locations modeled in terms of population size loss through time. Compared with the neutral landscape the Jean/Roach location lost 4.5% (p >0.05) and 8.6% (p >0.05) alleles/locus in current and future disturbance scenarios. This location

-5 lost 3.5% (p >0.05) and 10.3% (p = 2*10 ) Ho in current and future disturbance scenarios, relative to the neutral landscape. Only the neutral landscape exhibited stable genetic differentiation through time, with support for a cline. Spatial autocorrelation was present in the disturbance scenarios, with support for increased genetic clustering with little to no admixture (Fig. 3. 2. 7).

108

LAS VEGAS EAST

A population size loss of 28.9% was predicted in the current disturbance simulation, and

37.3% with future disturbance. In terms of loss of number of individuals, this location ranked 3RD (current disturbance) and 7TH (future disturbance), relative to the 17 locations modeled. Compared with the neutral landscape 8.1% (p >0.05) and 8.7% (p = 0.010) alleles/locus were lost in current and future disturbance scenarios. This location lost 5.6%

-5 (p = 0.003) and 7.2% (p = 2*10 ) Ho, relative to the neutral landscape in current and future disturbance scenarios. Genetic differentiation in the neutral landscape was predicted to rise through time as the result of rugged terrain; however, genetic differentiation increased to a lesser degree than in current or future disturbance scenarios.

Spatial autocorrelation remained in current and future disturbance scenarios with amplified population genetic structure caused by increased landscape barriers, resulting in populations with almost no admixture (Fig. 3. 2. 8).

LAS VEGAS NORTH

Current disturbance simulations predicted a 23.3% loss in population size, with future disturbance predicted at a 43.3% loss. Based on the 17 locations modeled, this location ranked 7TH (current disturbance) and 4TH (future disturbance) in terms of loss in population size. This location lost 5.7% (current disturbance, p = 0.002) and 10.8%

(future disturbance, p = 0.001) alleles/locus compared with the neutral landscape.

Heterozygosity was predicted to be lost by 2.6% (p = 0.010) and 8.2% (p = 1*10-5) in current and future disturbance scenarios, relative to the neutral landscape. In the neutral 109

landscape genetic differentiation increased and resulted in population genetic structure as the result of constricted habitat. As suitable habitat decreased with anthropogenic disturbance, spatial autocorrelation increased. Population genetic structure was predicted to be stronger with current disturbance, and strongest with future disturbance, which also exhibited increased genetic clustering (Fig. 3. 2. 9).

LAS VEGAS WEST

This location experienced expected population size losses of 40.9% with current disturbance, and 55.6% in future disturbance simulations. This location ranked 2ND in both current and future disturbance scenarios in terms of relative loss of number of individuals through time, relative to the 17 locations modeled. Compared with the neutral landscape this location lost 3.7% (p >0.05) and 0.8% (p >0.05) alleles/locus in current and future disturbance scenarios. The Las Vegas West location lost 10.8% (p = 3*10-7)

-13 and 30.0% (p = 4*10 ) Ho, relative to the neutral landscape in current and future disturbance scenarios. Only the neutral landscape showed indications of reaching steady genetic differentiation values. A natural corridor between rugged terrain created genetic structure and spatial autocorrelation was present in all landscape scenarios. Loss of suitable habitat within and surrounding the natural corridor in disturbance scenarios intensified population isolation, with notable reductions to gene flow (Fig. 3. 2. 10).

110

LAUGHLIN

A loss of 4.3% in number of individuals was predicted in the current disturbance simulation. A loss of 9.0% was predicted with future disturbance projections. This location ranked 16TH (current disturbance) and 17TH (future disturbance) in terms of loss in the population size, relative to the 17 locations modeled. The Laughlin location lost

1.1% (current disturbance, p >0.05) and 1.3% (future disturbance, p >0.05) alleles/locus compared with the neutral landscape. Heterozygosity was predicted to be lost by 0.4% (p

>0.05) and 0.9% (p = 0.002), relative to the neutral landscape, in current and future disturbance scenarios. The neutral landscape, current, and future disturbance scenarios all showed signs of achieving steady levels of genetic differentiation. Spatial autocorrelation was present in all landscape scenarios. Population genetic structure was weakly present in the neutral landscape as the result of natural landscape features in the northeast.

Anthropogenic disturbance strengthened this structure in both current and future scenarios. All landscapes are predicted to support admixture (Fig. 3. 2. 11).

MESQUITE

Population size losses of 11.9% and 15.7% were predicted with current and future disturbance simulations. In terms of population size loss through time, this location ranked 13TH in both current and future disturbance scenarios, based on the 17 locations modeled. Compared with the neutral landscape this location lost 1.9% (p >0.05) and

2.5% (p >0.05) alleles/locus in current and future disturbance scenarios. This location lost

1.0% (p >0.05) and 1.1% (p >0.05) Ho in current and future disturbance scenarios, 111

relative to the neutral landscape. The neutral landscape showed indications of reaching stable levels of genetic differentiation and exhibited spatial autocorrelation, with genetic structure analyses supporting two clusters, one on either side of the Virgin River. Both current and future disturbance scenarios increased population genetic structure.

Admixture was present in all scenarios, but decreased with disturbance (Fig. 3. 2. 12).

MOAPA VALLEY

A population size loss of 27.3% was predicted in the current disturbance simulation, with

43.9% in the future disturbance scenario. Based on the 17 locations modeled, the Moapa

Valley location ranked 4TH (current disturbance) and 3RD (future disturbance) in terms of loss in population size. Compared with the neutral landscape this location lost 5.3% (p

>0.05) and 9.6% (p >0.05) alleles/locus in current and future disturbance scenarios. In current and future disturbance scenarios Ho was predicted to be lost by 3.0% (p >0.05) and 7.0% (p = 1*10-6), relative to the neutral landscape. Genetic differentiation in the neutral landscape appeared stable and the current disturbance scenario reached steady levels over time. Spatial autocorrelation was present in all landscape scenarios.

Anthropogenic disturbance resulted in genetic structuring in both current and future scenarios (Fig. 3. 2. 13).

RED ROCK

This location experienced predicted population size losses of 63.6% with current disturbance, and 66.2% with future disturbance simulations. Red Rock ranked 1ST in 112

terms of population size loss in both current and future disturbance scenarios, relative to the 17 locations modeled. This location lost 13.5% (current disturbance, p = 7*10-4) and

5.1% (future disturbance, p = 2*10-4) alleles/locus compared with the neutral landscape.

Heterozygosity was predicted to be lost by 16.6% (p = 2*10-10) and 23.6% (p = 2*10-13) in current and future disturbance scenarios, relative to the neutral landscape. Only the neutral landscape exhibited steady genetic differentiation through time. Spatial autocorrelation was present in the neutral landscape with and weak genetic structure. As habitat amount was dramatically reduced by disturbance, isolation increased and gene flow was restricted (Fig. 3. 2. 14).

SANDY VALLEY

Population size losses of 4.2% and 13.3% were predicted in the current and future disturbance simulation. This location ranked 17TH (current disturbance) and 15TH (future disturbance) in terms of losses to the population size through time, relative to the 17 landscape locations modeled. Compared with the neutral landscape the Sandy Valley location lost 0.9% (p >0.05) and 3.0% (p >0.05) alleles/locus in current and future disturbance scenarios. This location lost 0.7% (p >0.05) and 3.2% (p >0.05) Ho, relative to the neutral landscape in current and future disturbance scenarios. Both of the disturbance landscapes showed indications of reaching steady genetic differentiation values through time. Spatial autocorrelation was present in all landscape scenarios, but increased with disturbance. In the neutral landscape, IBD was apparent with weak 113

population genetic structure. Anthropogenic disturbance amplified population genetic structure with admixture present; albeit reduced by disturbance (Fig. 3. 2. 15).

SEARCHLIGHT

A population size loss of 9.9% was predicted in the current disturbance simulation, and

13.5% with future disturbance. In terms of relative loss of the number of individuals through time, the Searchlight location ranked 14TH in both current and future disturbance scenarios, relative to the 17 locations modeled. This location lost 3.0% (current disturbance, p >0.05) and 3.2% (future disturbance, p >0.05) alleles/locus compared with

-4 the neutral landscape. This location lost 1.7% (p >0.05) and 2.2% (p = 2*10 ) Ho, relative to the neutral landscape in current and future disturbance scenarios. All three landscape scenarios appeared to reach stable levels of genetic differentiation. Spatial autocorrelation was present in the neutral landscape, along with weak population genetic structure. Spatial autocorrelation remained in current and future disturbance scenarios and population genetic structure increased. Admixture was predicted in all modeled landscapes; however, it was weakened by anthropogenic disturbance (Fig. 3. 2. 16).

TROUT CANYON

Current disturbance simulations predicted a 16.2% loss in number of individuals, decreasing by 18.1% with future disturbance. Based on the 17 locations modeled, this location ranked 11TH in terms of loss in population size in both current and future disturbance scenarios. Compared with the neutral landscape this location lost 3.8% (p 114

>0.05) and 4.7% (p = 0.036) alleles/locus in current and future disturbance scenarios.

-4 This location lost 1.8% (p >0.05) and 2.4% (p = 3*10 ) Ho, relative to the neutral landscape in current and future disturbance scenarios. Genetic differentiation in the neutral landscape was stable and both disturbance scenarios (current and future) showed indications of reaching stability. Spatial autocorrelation was not present in the neutral landscape, and genetic structure analyses revealed panmixia. Spatial autocorrelation formed in disturbance scenarios, with structure predicted to increase in the future disturbance scenario. All scenarios maintained gene flow, with decreases in disturbance landscapes (Fig. 3. 2. 17).

PREDICTED CONNECTIVITY SUCCESS

Current disturbance landscapes had 9 of 17 locations that maintained high levels of genetic connectivity (connectivity success index values ≥ 0.7) through time versus future disturbance scenarios, which had 5 of 17 (Table 3. 3). The Laughlin location ranked highest in genetic connectivity in both scenarios. Of the four landscapes that changed ranks from high levels of genetic connectivity between current and future scenarios two became intermediate (Coyote Springs, Sandy Valley) and two lost genetic connectivity

(Moapa Valley, Las Vegas North). Current and future disturbance landscapes had a comparable number of locations (3 and 4; respectively) with intermediate genetic connectivity (connectivity success index values 0.35 – 0.69). More locations (8 of 17) in future disturbance scenarios did not maintain genetic connectivity (connectivity success index values ≥ 0.34) through time compared with current disturbance landscapes (5 of 115

17). The Jean/Roach location ranked the lowest in both scenarios. Of the landscapes that failed to maintain genetic connectivity, 77% had negative connectivity success index values, meaning they fared worse than scenarios with a hypothetical absolute barrier, likely due to the compounding effects of habitat degradation and loss of individuals. No landscape increased in rank from current to future disturbance scenarios.

Landscape metrics by modeled location supported increased fragmentation and loss of suitable habitat with disturbance (Table 3. 4). The number of patches had a tendency to increase from the neutral landscape to the disturbed landscape (59% current and 76% future disturbance). Only the Red Rock location predicted more suitable habitat patches with the current landscape than the future scenario, likely because this location had more small suitable habitat patches with current disturbance. In the future disturbance scenario, the number of suitable habitat patches was reduced as they were converted to larger patches of unsuitable habitat. No landscapes experienced an increase in the largest suitable habitat patch size with disturbance. Conversely, there was a trend for unsuitable habitat to increase from the neutral landscape to current disturbance to future scenarios. The three exceptions were: Coyote Springs, where the largest unsuitable habitat patch remained constant; Jean/Roach, where there was no change between the neutral landscape and current disturbance; and Sandy Valley, where the largest unsuitable habitat patch did not change between current and future disturbance landscapes. These three landscapes shifted ranks in disturbance scenarios (lost genetic connectivity between scenarios) likely because habitat was degraded, but not completely lost. At all modeled locations suitable habitat decreased in transitioning from the neutral landscape to current 116

disturbance, as would be expected with the inclusion of disturbance. All modeled locations experienced further decreases in suitable habitat from current to future scenarios.

I found substantial overlap in landscape metric values when evaluated by connectivity success index rank (high, intermediate, and low/no genetic connectivity;

Table 3. 5). The predicted number of suitable habitat patches tended to increase as landscapes decreased in suitable habitat area, indicating an increased number of smaller patches in a more fragmented landscape. Neutral landscapes had no more than two patches of suitable habitat, while those with low/no genetic connectivity ranged from 2 –

13. Edge density was lowest in neutral landscapes (0 – 2.13 m/ha), indicating simplified landscape configurations. Landscapes that did not maintain genetic connectivity showed increasing complexity in configurations (2.38 – 3.98 m/ha). Edge density decreased at three locations (Ivanpah Valley, Moapa Valley, Red Rock) as the number of unsuitable habitat patches decreased and unsuitable habitat area increased in future disturbance scenarios; one (Red Rock) became dominated by unsuitable habitat. Largest patch index values indicated consistent, albeit variable, loss of area to the largest suitable habitat patch as genetic connectivity was lost. The largest patch of suitable habitat in neutral landscapes did not constitute less than 72% of the habitat, while the largest patch of unsuitable habitat was 0% – 23%. In landscapes that did not maintain genetic connectivity the largest suitable habitat patches ranged from 11% – 79%, while the largest unsuitable patches were up to 72%. Generally, percent area followed a similar pattern, with neutral landscapes having the most suitable habitat (72% - 100%). 117

Landscapes that failed to maintain genetic connectivity showed an increased amount of less suitable habitat (28% - 79%). The total percent core habitat by class also followed suite, with the largest area of suitable core habitat in neutral landscapes (53% - 85%). The smallest area of suitable habitat remained in landscapes with low/no genetic connectivity

(10% - 52%).

The highest ranking AIC models were those with number of suitable habitat patches and alleles/locus, percentage area of suitable habitat plus largest patch index and

Ho, and largest suitable habitat patch index and FST and N (Table 3. 6). Individual landscapes exhibited a general loss of genetic diversity, but these metrics tended to be less reliable indicators when evaluated across landscapes. Reductions in landscapes that maintained high levels of genetic diversity did not exceed 6%. Landscapes that allowed for intermediate gene flow were also not reduced by more than 6%. Low/no connectivity values tended to be higher; however, loss of genetic diversity could be as low as <1% in these landscapes. Because these estimates are heavily influenced by population size, care should be taken when interpreting results. Overall trends pointed to increased isolation with more complex landscape configuration as connectivity was lost and landscape composition was generally altered, decreasing dominance of the largest patch and amount of suitable habitat (Fig. 3. 3).

DISCUSSION 118

DISTURBANCE REDUCES GENE FLOW AND POPULATION SIZE

Habitat loss generally occurs continually; however, the relationship between habitat loss and population persistence is not linear and small additional losses near the threshold dramatically increase risk of local extinction (Fahrig 2001). In this study, I found that as anthropogenic disturbance increased, so did demographic and genetic effects. Habitat degradation resulted in population declines, with the most pronounced losses concomitant with increased habitat disturbance (given in the future prediction scenarios). For desert tortoises, genetic diversity was predicted to be highest in undisturbed (neutral) landscapes, and generally decreased as disturbance progressed. Additionally, the greater the disturbance the stronger the population genetic structure (see Fig. 3. 2 for results).

However, outcomes for maintaining genetic connectivity in disturbed landscapes are variable. While neutral landscapes always fared better than disturbance scenarios, the connectivity success index (based on genetic differentiation), predicted landscapes that maintain high levels of genetic connectivity experienced no more than a 27% loss in population. Losses as low as 12% were seen in landscapes with intermediate genetic connectivity. These landscapes generally did not lose more than 30% of their population.

Losses in low/no genetic connectivity landscapes were often higher, but could start at a

21% reduction. If these numbers hold true, this has significant implications given that

Mojave desert tortoises are estimated to have lost roughly 37% of their population range- wide from 2004 – 2014 (Allison & McLuckie 2018). 119

When genetic connectivity was high, population genetic structure showed substantial admixture when compared with neutral landscapes, while intermediate and low/no genetic connectivity landscapes lost gene flow. Genetic effects (i.e. loss of genetic diversity and increase in population genetic structure) indicate a landscape that lost connectivity within roughly five to 40 tortoise generations (85 – 680 years). Therefore, careful consideration of population declines and habitat development are needed to prevent deleterious effects to connectivity before they are apparent (Gregory & Beier

2014). In landscapes where increases in genetic structure and/or slight deviations in genetic diversity have been documented, efforts focused on reducing development pressures in tortoise habitat, restoring habitat, and adding permeability to landscape barriers will have the greatest benefit for connectivity.

CONNECTIVITY SUCCESS IS LANDSCAPE DEPENDENT

Given the challenges of capturing ecological processes with landscape patterns and the overlap I found when sorting landscapes by connectivity success index values, it is most appropriate to evaluate landscape locations as individual management units, rather than seek a single metric as a threshold. A single minimum for habitat amount is not practical, as the threshold will vary by landscape (Fahrig 2001). Disturbance landscapes with high levels of genetic connectivity tended towards low levels of landscape fragmentation and complexity. Suitable habitat amount and dominance of the largest patch remained high. It is important to note that even though high levels of gene flow were maintained in these disturbance landscapes, they all lost genetic connectivity, with 90% being the highest 120

level retained (Laughlin) and -124% being the lowest (Jean/Roach). Clearly, habitat loss and degradation are accelerated by development pressures. It is therefore not surprising that future disturbance scenarios had fewer landscapes that retained high levels of genetic connectivity (29%) and many (47%) failed to maintain genetic connectivity.

MANAGEMENT RECOMMENDATIONS

Balance between land use promoting economic and population growth and the long-term conservation and recovery of natural habitats and native species is the key purpose of the

Clark County, Nevada Multiple Species Habitat Conservation Plan (MSHCP 2000). My results indicate that current and planned activities related to economic and urban growth will result in desert tortoise population declines and loss of genetic connectivity, disrupting this balance. For the future development scenarios, data were not available regarding disturbance beyond the urban footprint (e.g. increased dirt roads). A literature review found anthropogenic disturbance to have a greater potential for habitat degradation than the development footprint alone (Lovich & Ennen 2011; Hunter et al.

2003). At each modeled location the total number of individuals was always highest with neutral landscape simulations and lowest with future projections of anthropogenic disturbance, with genetic diversity following the same pattern. Genetic differentiation and population genetic structure were always lowest in neutral landscapes and highest with future projections of disturbance. Therefore, I recommend more critical evaluation of proposed developments and reduction of anthropogenic disturbance in Mojave desert tortoise habitat. 121

Landscapes with higher levels of genetic connectivity should be prioritized for conservation to ensure additional habitat is not lost. Landscapes with intermediate genetic connectivity are excellent candidates for strategically restoring habitat and connectivity linkages. Models evaluating landscape change scenarios have shown that reductions in protected habitat results in large declines in connectivity, while corridors between protected areas may serve to increase connectivity (Cushman et al. 2016; Huxel &

Hastings 1999; Nowakowski et al. 2015). Protection of tortoise populations in landscape scenarios that fail to maintain genetic connectivity could benefit from major reductions in planned development and improvements to habitat in already disturbed areas. Ensuring

Mojave desert tortoise habitat is protected could move us towards reversing the trend of continually degrading habitat and reducing connectivity, and improve the opportunity for species recovery (Allison & McLuckie 2018; Averill-Murray et al. 2013; Boarman 2002) while preserving our unique natural heritage.

ACKNOWLEDGEMENTS

I am grateful to Scott Cambrin and Kimberley Jenkins (Clark County Desert

Conservation Program) for their support of this desert tortoise research (Desert Tortoise

Connectivity Modeling; 2015-UNR-1580A). Collaboration with Ken Nussear and Jill

Heaton made this project possible. They provided instruction, assistance, and support.

Amy Vandergast (USGS), Marjorie Matocq (UNR), and Todd Esque (USGS) provided noteworthy guidance and are valued colleagues. I also acknowledge the assistance of 122

Amy Vandergast and Anna Mitelberg (USGS) with genotyping and lab support. I thank

Lee Bice (Clark County Desert Conservation Program) for providing GIS layers. Many individuals contributed to initial data collection in the field including Kristina Drake,

Felicia Chen, Ben Gottsacker, Amanda McDonald, Jordan Swart, and Sara Murray. I appreciate their hard work and dedication. This project was funded by the Bureau of

Land Management through the sale of public lands as authorized by the Southern Nevada

Public Land Management Act. The Clark County Desert Conservation Program served as the funding source and point of contact. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the opinions or policies of the U.S. Government. Mention of trade names or commercial products does not constitute their endorsement by the U.S. Government. 123

REFERENCES

Allison LJ, McLuckie AM (2018) Population Trends in Mojave Desert Tortoises (Gopherus agassizii). Herpetological Conservation and Biology 13(2):433-452

Averill-Murray RC, Darst CR, Field KJ, Allison LJ (2012) A New Approach to Conservation of the Mojave Desert Tortoise. BioScience 62(10):893-899

Averill-Murray RC, Darst CR, Strout N, Wong M (2013) Conserving Population Linkages for the Mojave Desert Tortoise (Gopherus agassizii). Herpetological Conservation & Biology 8(1):1-15

Beier P, Loe S (1992) A Checklist for Evaluating Impacts to Wildlife Movement Corridors. Wildlife Society Bulletin 20(4):434-440

Beier P, Majka DR, Spencer WD (2008) Forks in the Road: Choices in Procedures for Designing Wildland Linkages. Conservation Biology 22(4):836-851

Beier P, Noss RF (1998) Do Habitat Corridors Provide Connectivity? Conservation Biology 12(6):1241-1252

Bivand R, Rundel C (2019) rgeos: An Interface to Geometry Engine – Open Source (‘GEOS’), R Package Version 0.4-3. https://CRAN.R-project.org/package=rgeos

BLM (Bureau of Land Management) (2018a) Programs: Energy and Minerals Renewable Energy Data Table Solar Historical Block – Solar Energy Projects Approved since 2009 https://www.blm.gov/basic/programs-energy-and-minerals-renewable-energy-data-table- solar-historical-block Accessed June 2019 Accessed June 2019

124

BLM (Bureau of Land Management) (2018b) Solar Energy Projects Approved www.blm.gov/sites/blm.gov/files/energy_renewable_SolarProjectInfo_november%20201 8%20%282%29_0.xlsx Accessed June 2019

BLM (Bureau of Land Management) (2018c) Pending Solar Energy Projects www.blm.gov/sites/blm.gov/files/energy_renewable_SolarProjectInfo_1.xlsx Accessed June 2019

Boarman WI (2002) Threats to Desert Tortoise Populations: A Critical Review of the Literature. US Geological Survey, Western Ecological Research Center

Boarman WI, Sazaki M (2006) A Highway’s Road-Effect Zone for Desert Tortoises (Gopherus agassizii). Journal of Arid Environments 65(1):94-101

Britten HB, Riddle BR, Brussard PF, Marlow R, Lee TE Jr (1997) Genetic Delineation of Management Units for the Desert Tortoise, Gopherus agassizii, in Northeastern Mojave Desert. Copeia 3:523-530

Carter SK, Nussear KE, Esque TC, Leinwand IF, Masters E, Inman RD, Carr NB, Allison LJ (in review) Connecting Landscape Approaches and Species-Level Conservation: Quantifying Development to Inform Management of Mojave and Sonoran Desert Tortoise Habitat. Endangered Species Research

Chetkiewicz CLB, St. Clair CC, Boyce MS (2006) Corridors for Conservation: Integrating Pattern and Process. Annual Review of Ecology, Evolution, and Systematics 37:317-342

Clark County (2017) 2018 Biennial Adaptive Management Report. Prepared by: Alta Science and Engineering Inc., Desert Conservation Program, Clark County, Nevada 125

Creech TG, Epps CW, Landguth EL, Wehausen JD, Crowhurst RS, Holton B, Monello RJ (2017) Simulating the Spread of Selection-Driven Genotypes using Landscape Resistance Models for Desert Bighorn Sheep. PLoS ONE 12(5):e0176960

Cushman SA, Elliot NB, Macdonald DW, Loveridge AJ (2016) A Multi-Scale Assessment of Population Connectivity in African Lions (Panthera leo) in Response to Landscape Change. Landscape Ecology 31(6):1337-1353

Cushman SA, McGarigal K, Maile CN (2008) Parsimony in Landscape Metrics: Strength, Universality, and Consistency. Ecological Indicators 8(5):691-703

Darst CR, Murphy PJ, Strout NW, Campbell SP, Field KJ, Allison L, Averill-Murray RC (2013) A Strategy for Prioritizing Threats and Recovery Actions for At-Risk Species. Environmental Management 51(3):786-800

Doak D, Kareiva P, Klepetka B (1994) Modeling Population Viability for the Desert tortoise in the Western Mojave Desert. Ecological Applications 4(3):446-460

Dutcher KE, Vandergast AG, Esque TC, Mitelberg A, Matocq MD, Heaton JS, Nussear KE (2020) Genes in Space: What Mojave Desert Tortoise Genetics Can Tell Us About 126

Landscape Connectivity. Conservation Genetics https://doi.org/10.1007/s10592-020- 01251-z

Edwards T, Lathrop A, Ngo A, Choffe K, Murphy RW (2003) STR/Microsatellite Primers for the Desert Tortoise, Gopherus agassizii, and its Congeners. Conservation Genetics Resources 3(2):365-368

Evanno G, Regnaut S, Goudet J (2005) Detecting the Number of Clusters of Individuals using the Software STRUCTURE: A Simulation Study. Molecular Ecology 14(8):2611- 2620

Ewers RM, Didham RK (2006) Confounding Factors in the Detection of Species Responses to Habitat Fragmentation. Biological Reviews 81(1):117-142

Fahrig L (2001) How Much Habitat is Enough? Biological Conservation 100(1):65-74

Fahrig L (2003) Effects of Habitat Fragmentation on Biodiversity. Annual Review of Ecology, Evolution and Systematics 34(1):487-515

127

Francis RM (2017) PopHelper: An R Package and Web App to Analyze and Visualize Population Structure. Molecular Ecology Resources 17(1):27-32

Gauffre B, Mallez S, Chapuis MP, Leblois R, Litrico I, Delaunay S, Badenhausser I (2015) Spatial Heterogeneity in Landscape Structure Influences Dispersal and Genetic Structure: Empirical Evidence from a Grasshopper in an Agricultural Landscape. Molecular Ecology 24(8):1713-1728

Gonzales E, Hamrick JL, Smouse PE, Trapnell DW, Peakall R (2010) The Impact of Landscape Disturbance on Spatial Genetic Structure in the Guanacaste Tree, Enterolobium cyclocarpum (Fabaceae). Journal of Heredity 101(2):133-143

Gregory AJ, Beier P (2014) Response Variables for Evaluation of the Effectiveness of Conservation Corridors. Conservation Biology 28(3):689-695

Goudet J (2005) Hierfstat, a Package for R to Compute and test Variance Components and F-Statistics. Molecular Ecology Notes 5:184-186

Haddad NM, Holt RD, Fletcher RJ Jr, Loreau M, Clobert J (2017) Connecting Models, Data, and Concepts to Understand Fragmentation’s Ecosystem-Wide Effects. Ecography 40(1):1-8

128

Hagerty BE, Nussear KE, Esque TC, Tracy CR (2011) Making Molehills out of Mountains: Landscape Genetics of the Mojave Desert Tortoise. Landscape Ecology 26(2):267-280

Hagerty BE, Peacock MM, Kirchoff VS, Tracy CR (2008) Polymorphic Microsatellite Markers for the Mojave Desert Tortoise, Gopherus agassizii. Molecular Ecology Resources 8(5):1149-1151

Hagerty BE, Tracy CR (2010) Defining Population Structure for the Mojave Desert Tortoise. Conservation Genetics 11(5):1795-1807

Hansen AJ, DeFries R (2007) Ecological Mechanisms Linking Protected Areas to Surrounding Lands. Ecological Applications 17(4):974-988

Hesselbarth MH, Sciaini M, With KA, Wiegand K, Nowosad (2019) landscapemetrics: An Open-Source R Tool to Calculate Landscape Metrics. Ecography 42(10):1648-1657

Hijmans RJ (2019) raster: Geographic Data Analysis and Modeling, R Package Version 2.9-5. https://CRAN.R-project.org/package=raster

Hunter LM, Gonzalez G. MJ, Stevenson M, Karish KS, Toth R, Edwards Jr. TC, Lilieholm RJ, Cablk M (2003) Population and Land Use Change in the California Mojave: Natural Habitat Implications of Alternative Futures. Population Research and Policy Review 22(4):373-397 129

Huxel GR, Hastings A (1999) Habitat Loss, Fragmentation, and Restoration. Restoration Ecology 7(3):309-315

Inman RD, Esque TC, Nussear KE, Leitner P, Matocq MD, Weisberg PJ, Dilts TE, Vandergast AG (2013) Is there Room for All of Us? Renewable Energy and Xerospermophilus mohavensis. Endangered Species Research 20(1):1-18

Jombart T (2008) adegenet: A R Package for the Multivariate Analysis of Genetic Markers. Bioinformatics 24(11):1403-1405

Jombart T, Dufour AB, Pontier D (2008) Revealing Cryptic Spatial Patterns in Genetic Variability by a New Multivariate Method. Heredity 101(1):92-103

Kahle D, Wickman H (2013) ggmap: Spatial Visualization with ggplot2. The R Journal 5(1):144-161

Kupfer JA (2012) Landscape Ecology and Biogeography: Rethinking landscape Metrics in a Post-Fragstats Landscape. Progress in Physical Geography 36(3):400-420

Landguth EL, Cushman SA, Schwartz MK, McKelvey KS, Murphy M, Luikart G (2010) Quantifying the Lag Time to Detect Barriers in Landscape Genetics. Molecular Ecology 19(19):4179-4191

Leblois R, Rousset F, Estoup A (2004) Influence of Spatial and Temporal Heterogeneities on the Estimation of Demographic Parameters in a Continuous Population Using Individual Microsatellite Data. Genetics 166(2):1081-1092

130

Lovich JE, Ennen JR (2011) Wildlife Conservation and Solar Energy Development in the Desert Southwest, United States. BioScience 61(12):982-992

McRae BH, Beier P, Dewald E, Huynh Y, Keim P (2005) Habitat Barriers Limit Gene Flow and Illuminate Historical Events in a Wide-Ranging Carnivore, the American Puma. Molecular Ecology 14(7):1965-1977

Morafka DJ, Berry KH (2002) Is Gopherus agassizii a Desert-Adapted, or an Exaptive Opportunist? Implications for Tortoise Conservation. Chelonian Conservation and Biology 4(2):263-287

MSHCP (2000) Final Clark County Multiple Species Habitat Conservation Plan and Environmental Impact Statement for Issuance of a Permit to Allow Incidental Take of 79 Species in Clark County, Nevada. Clark County Department of Comprehensive Planning, Las Vegas, Nevada

Murphy MA, Evans JS, Cushman SA, Storfer A (2008) Representing Genetic Variation as Continuous Surfaces: An Approach for Identifying Spatial Dependency in Landscape Genetic Studies. Ecography 31(6):685-697

131

Nei M (1973) Analysis of Gene Diversity in Subdivided Populations. In: Proceedings of the National Academy of Sciences 70(12):3321-3323

Noss RF (1987) Corridors in Real Landscapes: A Reply to Simberloff and Cox. Conservation Biology 1(2):159-164

Nowakowski AJ, Dewoody JA, Fagan ME, Willoughby JR, Donnelly MA (2015) Mechanistic Insights into Landscape Genetic Structure of Two Tropical Amphibians using Field-Derived Resistance Surface. Molecular Ecology 24(3):580-595

O’Connor MP, Zimmerman, LC, Ruby DE, Bulova SJ, Spotila JR (1994) Home Range Size and Movements by Desert Tortoises, Gopherus agassizii, in the Eastern Mojave Desert. Herpetological Monographs 8:60-71

Pendleton BK, Chambers JC, Brooks ML, Ostoja SM (2013) Ecosystem Stressors in Southern Nevada. USDA Forest Service Technical Report RMRS-GTR-303

Pritchard J K, Stephens M, Donnelly P (2000) Inference of Population Structure using Multilocus Genotype Data. Genetics 155(2):945-959

R Core Team (2019) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/

132

Rebaudo F (2014) SIMADAPT (Version 1.8.0). CoMSES Computational Model Library https://www.comses.net/codebases/3137/releases/1.8.0/

Row JR, Brooks RJ, Mackinnon CA, Lawson A, Crother BI, White M, Lougheed SC (2011) Approximate Bayesian Computation Reveals the Factors that Influence Genetic Diversity and Population Structure of Foxsnakes. Journal of Evolutionary Biology 24(11):2364-2377

Sanchez-Ramirez S, Rico Y, Berry K, Edwards T, Karl AE, Henen BT, Murphy RW (2018) Landscape Limits Gene Flow and Drives Population Structure in Agassiz’s Desert Tortoise (Gopherus agassizii). Scientific Reports 8(1):1-17

Segelbacher G, Hoglund J, Storch I (2003) From Connectivity to Isolation: Genetic Consequences of Population Fragmentation in Capercaillie across Europe. Molecular Ecology 12(7):1773-1780

133

Thatte P, Joshi A, Vaidyanathan S, Landguth E, Ramakrishnan U (2018) Maintaining Tiger Connectivity and Minimizing Extinction into the Next Century: Insights for Landscape Genetics and Spatially Explicit Simulations. Biological Conservation 218:181-191

USCB (US Census Bureau) (2018) Fastest Growing Cities and Fastest Growing States www.census.gov Accessed June 2019

USFWS (US Fish & Wildlife Service) (1994) Desert Tortoise (Mojave Population) Recovery Plan. US Fish & Wildlife Service, Portland, OR

USFWS (US Fish & Wildlife Service) (2011) Revised Recovery Plan for the Mojave Population of the Desert Tortoise (Gopherus agassizii). Pacific Southwest Region, Sacramento, CA

Vandergast AG, Bohonak AJ, Weissman DB, Fisher RN (2007) Understanding the Genetic Effects of Recent Habitat Fragmentation in the Context of Evolutionary History Phylogeography and Landscape Genetics of a Southern California Endemic Jerusalem Crickey (Orthoptera: Stenopelmatidae: Stenopelmatus). Molecular Ecology 16(5):977- 992 134

vonSekendorff Hoff K, Marlow RW (2002) Impacts of Vehicle Road Traffic on Deseret Tortoise Populations with Consideration of Tortoise Habitat in Southern Nevada. Chelonian Conservation and Biology 4(2):449-456

Waples R, Gaggiotti O (2006) What is a Population? An Empirical Evaluation of Some Genetic Methods for Identifying the Number of Gene Pools and Their Degree of Connectivity. Publication, Agencies and Staff of the US Department of Commerce 463 135

TABLES AND FIGURES

Table 3. 1. Conversion factors (CF) used to adjust habitat suitability values (HSV) for models with anthropogenic disturbance. The inverse of habitat suitability was used to calculate landscape resistance.

Example HSV Scaled Resistance Disturbance Type CF HSV by CF Value

None 0 0.500 0.500 0.500

Urban/Cleared Land 1.00 0.500 0 1.000

Solar Energy 1.00 0.500 0 1.000

Railways 0.75 0.500 0.125 0.875

Major Roads 0.75 0.500 0.125 0.875

Minor Roads (max length) 0.25 0.500 0.375 0.625

Right-of-Ways 0.25 0.500 0.375 0.625

136

Table 3. 2. Results for each landscape scenario: neutral; current disturbance; future disturbance based on a 50 year forecast. Reported values use average outcomes ± standard deviation. Results for: (N) number of individuals; (A) mean number of alleles/locus; (Ho) observed heterozygosity; (FST) pairwise genetic differentiation; (K) number of genetic clusters. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1.

N A Ho FST K

Boulder City

Neutral 9889.8 ± 34.7 23.7 ± 0.5 0.797 ± 0.007 0.001 ± < 0.001 1

Current 7644.6 ± 26.5 22.5 ± 0.5 0.774 ± 0.008 0.004 ± < 0.001 2

Future 5847.0 ± 16.2 21.5 ± 0.5 0.734 ± 0.009 0.008 ± 0.001 2

Coyote Springs

Neutral 7973.8 ± 23.7 22.8 ± 0.4 0.781 ± 0.006 0.003 ± <0.001 2

Current 6892.6 ± 29.9 22.2 ± 0.4 0.765 ± 0.010 0.006 ± 0.001 2

Future 6612.1 ± 20.3 22.1 ± 0.4 0.761 ± 0.009 0.008 ± 0.001 2

Dry Lake

Neutral 11348.1 ± 20.5 24.3 ± 0.4 0.804 ± 0.006 0.001 ± < 0.001 1

Current 8604.9 ± 19.6 23.1 ± 0.4 0.778 ± 0.009 0.005 ± 0.001 2

Future 7923.7 ± 27.8 22.8 ± 0.5 0.771 ± 0.006 0.006 ± 0.001 3

137

Eldorado Valley

Neutral 7919.5 ± 23.3 22.7 ± 0.5 0.788 ± 0.006 0.002 ± < 0.001 2

Current 7298.1 ± 25.7 22.3 ± 0.4 0.782 ± 0.007 0.003 ± < 0.001 2

Future 6992.5 ± 26.4 21.9 ± 0.5 0.771 ± 0.010 0.004 ± < 0.001 2

Indian Springs

Neutral 6370.7 ± 18.7 21.9 ± 0.4 0.774 ± 0.007 0.002 ± < 0.001 2

Current 4955.9 ± 25.3 21.0 ± 0.4 0.743 ± 0.010 0.005 ± 0.001 2

Future 4860.5 ± 29.7 20.7 ± 0.5 0.740 ± 0.009 0.005 ± 0.001 2

Ivanpah Valley

Neutral 7847.4 ± 25.5 22.8 ± 0.5 0.778 ± 0.007 0.003 ± < 0.001 2

Current 5984.4 ± 22.2 21.4 ± 0.5 0.746 ± 0.010 0.010 ± 0.001 5

Future 5523.5 ± 20.9 21.3 ± 0.5 0.741 ± 0.008 0.016 ± 0.002 2

Jean/Roach

Neutral 8723.0 ± 25.4 23.2 ± 0.5 0.791 ± 0.006 0.002 ± < 0.001 2

Current 6869.8 ± 24.8 22.2 ± 0.5 0.764 ± 0.007 0.010 ± 0.001 2

Future 5320.7 ± 19.4 21.2 ± 0.6 0.710 ± 0.011 0.023 ± 0.003 3

Las Vegas East

Neutral 6924.5 ± 29.5 22.1 ± 0.5 0.759 ± 0.010 0.010 ± 0.001 2

Current 4924.0 ± 25.0 20.3 ± 0.5 0.717 ± 0.010 0.024 ± 0.003 2

Future 4338.3 ± 26.7 20.2 ± 0.4 0.704 ± 0.008 0.027 ± 0.003 2

138

Las Vegas North

Neutral 7029.6 ± 20.2 22.3 ± 0.4 0.783 ± 0.006 0.006 ± 0.001 2

Current 5390.3 ± 25.7 21.0 ± 0.4 0.762 ± 0.008 0.007 ± 0.001 2

Future 3987.8 ± 16.5 19.9 ± 0.5 0.718 ± 0.010 0.012 ± 0.001 8

Las Vegas West

Neutral 6653.6 ± 22.2 22.1 ± 0.5 0.759 ± 0.009 0.009 ± 0.001 2

Current 3929.4 ± 18.9 21.3 ± 0.5 0.677 ± 0.008 0.020 ± 0.002 2

Future 2951.7 ± 14.3 21.9 ± 0.4 0.531 ± 0.012 0.037 ± 0.006 2

Laughlin

Neutral 8992.1 ± 30.0 23.2 ± 0.2 0.788 ± 0.006 0.002 ± < 0.001 2

Current 8603.4 ± 25.6 22.9 ± 0.4 0.785 ± 0.006 0.003 ± < 0.001 3

Future 8180.8 ± 20.5 22.9 ± 0.5 0.781 ± 0.007 0.003 ± < 0.001 3

Mesquite

Neutral 9928.4 ± 20.8 23.6 ± 0.6 0.790 ± 0.007 0.005 ± 0.001 2

Current 8746.8 ± 22.5 23.1 ± 0.5 0.782 ± 0.007 0.008 ± 0.001 2

Future 8365.3 ± 22.6 23.0 ± 0.5 0.782 ± 0.005 0.008 ± 0.001 2

Moapa Valley

Neutral 10951.7 ± 20.8 24.1 ± 0.4 0.801 ± 0.005 0.002 ± < 0.001 1

Current 7959.9 ± 23.5 22.8 ± 0.5 0.777 ± 0.008 0.004 ± 0.001 2

Future 6139.5 ± 23.9 21.8 ± 0.5 0.746 ± 0.008 0.009 ± 0.001 2

139

Red Rock

Neutral 7295.1 ± 23.2 22.4 ± 0.3 0.777 ± 0.008 0.002 ± < 0.001 2

Current 2653.0 ± 15.1 19.3 ± 0.5 0.648 ± 0.014 0.026 ± 0.004 2

Future 2468.0 ± 17.4 21.2 ± 0.4 0.594 ± 0.011 0.028 ± 0.003 2

Sandy Valley

Neutral 6574.8 ± 26.8 21.9 ± 0.6 0.771 ± 0.010 0.004 ± < 0.001 2

Current 6298.1 ± 29.4 21.7 ± 0.4 0.766 ± 0.008 0.006 ± 0.001 2

Future 5703.0 ± 21.8 21.2 ± 0.5 0.746 ± 0.010 0.011 ± 0.002 2

Searchlight

Neutral 10144.9 ± 17.2 23.8 ± 0.4 0.797 ± 0.008 0.002 ± < 0.001 1

Current 9137.6 ± 22.3 23.1 ± 0.5 0.783 ± 0.008 0.004 ± 0.001 3

Future 8775.4 ± 23.0 23.0 ± 0.5 0.779 ± 0.007 0.004 ± 0.001 3

Trout Canyon

Neutral 10158.2 ± 20.3 23.7 ± 0.6 0.795 ± 0.006 0.001 ± < 0.001 1

Current 8516.9 ± 23.4 22.8 ± 0.5 0.781 ± 0.007 0.003 ± < 0.001 3

Future 8314.7 ± 20.7 22.6 ± 0.5 0.777 ± 0.007 0.004 ± 0.001 3

140

Table 3. 3. Connectivity success index (CSI) values for current disturbance (CD) and future disturbance (FD) based on a 50 year forecast. Predicted FST values were used in the CSI. Those near 1 indicate gene flow comparable to neutral landscapes; those near or below 0 indicate failure to maintain connectivity. High and low values in bold.

Location CSI (CD) CSI (FD)

Boulder City 0.61 -0.05

Coyote Springs 0.74 0.55

Dry Lake 0.55 0.44

Eldorado Valley 0.89 0.84

Indian Springs 0.82 0.80

Ivanpah Valley 0.32 -0.25

Jean/Roach -50.84 -124.30

Las Vegas East -0.90 -1.30

Las Vegas North 0.80 0.28

Las Vegas West -0.18 -2.12

Laughlin 0.90 0.89

Mesquite 0.55 0.41

Moapa Valley 0.72 0.19

Red Rock -0.98 -1.11

Searchlight 0.85 0.80

Sandy Valley 0.85 0.44

Trout Canyon 0.88 0.81 141

Table 3. 4. Landscape metrics for suitable and unsuitable habitat by modeled locations in neutral landscapes, (CD) current disturbance, and (FD) future disturbance projections. Ranks were established using connectivity success index (CSI) values: (1) high genetic connectivity; (2) intermediate genetic connectivity; (3) low/no genetic connectivity. Landscape metrics: (NP) number of patches; (LPI) largest patch index percentage; (A2) percentage area of landscape.

Ranks Neutral CD FD

CD FD NP LPI A2 NP LPI A2 NP LPI A2

Boulder City 2 3

Suitable Habitat 1 97.8 97.8 2 82.2 86.0 4 59.3 67.5

Unsuitable Habitat 2 1.2 2.2 11 7.2 14.0 6 28.3 32.5

Coyote Springs 1 3

Suitable Habitat 1 86.1 86.1 1 82.2 82.2 2 72.6 77.9

Unsuitable Habitat 4 10.6 13.9 7 10.6 17.8 7 10.6 22.1

Dry Lake 2 2

Suitable Habitat 1 99.0 99.0 2 66.4 87.7 3 61.0 82.2

Unsuitable Habitat 1 1.0 1.0 9 8.6 12.3 5 14.4 17.8

Eldorado Valley 1 1

Suitable Habitat 1 83.2 83.2 1 81.8 81.8 1 81.1 81.1

Unsuitable Habitat 4 15.2 16.8 6 15.4 18.2 9 15.4 18.9 142

Indian Springs 1 1

Suitable Habitat 1 72.3 72.3 3 65.2 65.8 3 63.5 64.2

Unsuitable Habitat 5 23.0 27.7 10 25.2 34.2 9 26.3 35.8

Ivanpah Valley 3 3

Suitable Habitat 2 83.5 83.7 3 72.5 72.8 3 36.5 65.6

Unsuitable Habitat 8 7.0 16.3 14 8.5 27.2 9 20.3 34.4

Jean/Roach 3 3

Suitable Habitat 1 87.5 87.5 2 79.2 79.4 3 49.1 63.0

Unsuitable Habitat 2 9.9 12.5 7 9.9 60.6 5 22.2 37.0

Las Vegas East 3 3

Suitable Habitat 1 97.5 97.5 4 42.5 75.4 5 35.2 66.5

Unsuitable Habitat 5 1.3 2.5 6 20.0 5 21.9 33.5

Las Vegas North 1 3 24.6

Suitable Habitat 1 85.9 85.9 1 73.3 73.3 5 53.0 54.9

Unsuitable Habitat 4 11.0 14.1 10 11.0 26.7 4 16.2 45.1

Las Vegas West 3 3

Suitable Habitat 1 81.3 81.3 4 44.8 46.0 5 11.3 32.2

Unsuitable Habitat 8 13.2 18.7 8 36.3 54.0 4 67.3 67.8

Laughlin 1 1

Suitable Habitat 1 97.3 97.3 1 92.5 92.5 1 89.4 89.4

Unsuitable Habitat 3 1.3 2.7 7 2.6 7.5 8 3.2 10.6 143

Mesquite 2 2

Suitable Habitat 1 95.0 95.0 2 91.2 91.4 3 85.6 86.4

Unsuitable Habitat 7 1.8 2.7 7 4.6 8.6 4 8.2 13.6

Moapa Valley 1 3

Suitable Habitat 1 100 100 1 85.0 85.0 3 44.5 64.2

Unsuitable Habitat 0 0 0 10 10.9 15.0 5 33.9 35.8

Red Rock 3 3

Suitable Habitat 1 92.0 92.0 13 13.7 36.0 6 12.8 27.5

Unsuitable Habitat 13 2.2 8.0 4 63.3 64.0 2 72.2 72.5

Sandy Valley 1 2

Suitable Habitat 2 80.2 80.5 2 76.5 76.7 2 69.7 69.8

Unsuitable Habitat 6 14.3 19.5 9 15.8 23.3 8 15.8 30.2

Searchlight 1 1

Suitable Habitat 1 99.0 99.0 2 90.9 93.6 2 88.2 90.7

Unsuitable Habitat 3 0.6 1.0 6 3.8 6.4 10 5.9 9.3

Trout Canyon 1 1

Suitable Habitat 1 89.4 89.4 1 87.0 87.0 1 85.0 85.0

Unsuitable Habitat 1 10.6 10.6 4 10.6 13.0 6 10.6 15.0

144

Table 3. 5. Landscape metrics of suitable and unsuitable habitat by ability to maintain genetic connectivity. Assignments were based on ranking connectivity success index (CSI) values for disturbed landscapes: high ≥ 0.70, intermediate = 0.35-0.69, low/no genetic connectivity ≤0.34. Landscape metrics: (NP) number of patches; (LPI) largest patch index percentage; (A2) percentage area of landscape; (Core) percentage of core area landscape; (Edge) edge density in m/ha. Edge values are equivalent by category.

NP LPI A2 Core Edge

Neutral

Suitable Habitat 1-2 72.3-100 72.3-100 53.3-84.6 0-2.1

Unsuitable Habitat 0-13 0-23 0-27.7 0-14.5 0-2.1

High

Suitable Habitat 1-3 63.5-92.5 64.2-93.6 35.7-71.0 1.0-3.2

Unsuitable Habitat 4-10 2.6-26.3 6.4-35.8 0.2-15.0 1.0-3.2

Intermediate

Suitable Habitat 2-3 61.0-91.2 69.8-91.4 41.0-68.0 1.6-3.1

Unsuitable Habitat 4-11 4.6-15.8 8.6-30.2 1.3-10.0 1.6-3.1

Low/No

Suitable Habitat 2-13 11.3-79.2 27.5-79.4 9.8-52.0 2.4-4.0

Unsuitable Habitat 2-14 8.5-72.2 20.6-72.5 4.2-48.8 2.4-4.0

145

Table 3. 6. Akaike’s information criterion (AIC) ranks of the strength of relationships. Terms are metrics of landscape disturbance with population or genetic statistics.

Population and genetic statistics: (N) number of individuals; (A) alleles/locus; (Ho) observed heterozygosity; (FST) genetic differentiation. Landscape metrics: (NP) number of suitable habitat patches; (LPI) largest suitable habitat patch index percentage; (A2) percentage area of suitable habitat. Highest ranking models are bolded statistic.

Model AIC for N AIC for FST AIC for Ho AIC for A

LPI 557.5 -289.2 -149.3 64.4

LPI + A2 559.4 -287.4 -158.0 64.6

NP 567.5 -259.9 -124.1 58.6

LPI + NP 558.7 -287.8 -147.3 60.5

LPI + NP + A2 560.5 -285.9 -156.2 58.9

NP + A2 564.4 -274.5 -155.5 60.0

A2 565.8 -272.7 -157.5 69.7

146

Fig. 3. 1. Resistance surfaces of study landscape (Clark County, Nevada) with a 20 km buffer. Left to right: neutral landscape without disturbance; current disturbance; future projections of disturbance based on a 50 year forecast. 147

Fig. 3. 2. 1. Boulder City Conservation Easement North (600 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A)

Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 148

Fig. 3. 2. 2. Coyote Springs (625 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 149

Fig. 3. 2. 3. Dry Lake (625 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 150

Fig. 3. 2. 4. Eldorado Valley (625 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 151

Fig. 3. 2. 5. Indian Springs (600 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 152

Fig. 3. 2. 6. Ivanpah Valley (625 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 153

Fig. 3. 2. 7. Jean/Roach (625 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 154

Fig. 3. 2. 8. Las Vegas East (525 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 155

Fig. 3. 2. 9. Las Vegas North (525 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 156

Fig. 3. 2. 10. Las Vegas West (600 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 157

Fig. 3. 2. 11. Laughlin (625 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 158

Fig. 3. 2. 12. Mesquite (625 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 159

Fig. 3. 2. 13. Moapa Valley (625 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 160

Fig. 3. 2. 14. Red Rock (600 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 161

Fig. 3. 2. 15. Sandy Valley (600 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 162

Fig. 3. 2. 16. Searchlight (625 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 163

Fig. 3. 2. 17. Trout Canyon. (625 km2). Upper (left to right): location within Clark County, Nevada; neutral representation without disturbance; current disturbance; future disturbance based on a 50 year forecast; hypothetical absolute barrier. Lower (left to right): simulation results for neutral, current, and future disturbance. A) Time-series: number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) sPCA genetic patterns; C) STRUCTURE barplots. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1. 164

165

Fig. 3. 3. Landscape metrics of disturbance relative to population and genetic statistics. Each figure compares a landscape metric on the x-axis to the increase in number of individuals (N), alleles/locus (A), observed heterozygosity (Ho), and genetic differentiation (FST) on the y-axis. Top to bottom: A) number of suitable habitat patches

(along the x-axis) where N, A, and Ho are reduced and FST is increased with greater patchiness on the landscape; B) largest suitable habitat patch index percentage (along the x-axis) where N, A, and Ho are reduced and FST is increased as suitable habitat patches decrease in size; C) percentage suitable habitat area (along the x-axis) where N, A, and Ho are reduced and FST is increased as suitable habitat area decreases. 166

APPENDIX A: SUPPLEMENTAL MATERIAL FOR CHAPTER 1

Table A 1. Summary of primer and locus information to genotype Gopherus agassizii.

# OF SIZE GENBANK LOCUS REPEAT MOTIF REFERENCE ALLELES RANGE #

GOA1 5 128 - 143 (CATC)7 EU285463 Hagerty et al. 2008

(GCT)6(GAGCAC GOA2 4 204 - 213 EU285462 Hagerty et al. 2008 TAGGACCTC)

(GCT)5(CTTCGT)

GOA3 9 285 - 309 (CAT)7 AY317141 Edwards et al. 2003 (GCT)2

GOA4 13 264 - 303 (CAT)11 UE285460 Hagerty et al. 2008

GOA6 17 213 - 277 (CTAT)10 EU285464 Hagerty et al. 2008

GOA8 17 146 - 211 (TAGA)16 EU285470 Hagerty et al. 2008

(CATC)19(CCAT) GOA9 21 206 - 290 EU285468 Hagerty et al. 2008 (CTAT)8

GOA11 22 268 - 341 (CAT)8 EU285458 Hagerty et al. 2008 167

GOA12 17 105 - 170 (CAT)37 EU285461 Hagerty et al. 2008

GOA13 16 208 - 296 (CTAT)11 EU285465 Hagerty et al. 2008

GOA14 43 202 - 310 (TAGA)12 EU285471 Hagerty et al. 2008

GOA17 26 214 - 319 (CTAT)16 EU285466 Hagerty et al. 2008

GOA22 18 152 - 221 (CTAT)18 EU285467 Hagerty et al. 2008

GOA23 18 175 - 262 (TAGA)18 EU285469 Hagerty et al. 2008

GOAG4 19 140 - 199 (CAA)24 AY317142 Edwards et al. 2003

GOAG7 5 266 - 280 (AC)3(GC)5(AC)11 AY317145 Edwards et al. 2003

GP30 9 203 - 240 (GT)13 AF546889 Schwartz et al. 2003

GP55 9 272 - 307 (GT)9 AF546893 Schwartz et al. 2003 168

GP61 23 203 - 258 (GT)12 AF546896 Schwartz et al. 2003

GP81 7 377 - 397 (GT)11(GA)10 AF546894 Schwartz et al. 2003

169

APPENDIX B: SUPPLEMENTAL MATERIAL FOR CHAPTER 2

Statistics derived from a genetic dataset, with samples from a continuous population in the Ivanpah Valley, was used to parameterize genotypes for model behavior and computational limitations simulations.

MODEL BEHAVIOR

The number of individuals in the neutral landscape ranged from 8750 – 8412 at generations 0 and 200, decreasing 3.9%. Landscapes with an absolute barrier (horizontal or vertical) decreased to N = 8026 (8.3%) by generation 200. The mean number of alleles/locus at generations 0 and 200 ranged from 11.0 - 9.5 in the neutral landscape and was similar in barrier models. Heterozygosity decreased initially in all models from 0.818 to 0.813 but was relatively stable through time. By generation 200 observed heterozygosity (Ho) ranged from 0.762 in the neutral landscape to 0.750 and 0.758 in barrier models.

Genetic differentiation was lowest in the neutral landscape (FST = 0.0006 –

0.0020) with barrier models increasing through time (FST = 0.0120 and 0.0090 by generation 200). Population genetic structure at generation five indicated no spatial autocorrelation in any model; however, by generation 200 there was strong spatial autocorrelation in barrier models with clusters structured on either side of the linear feature. This was further evidenced by STRUCTURE, which calculated Pr(X|K) = 1 and ΔK 170

= 2 in neutral models, and K = 2 (Pr(X|K) and ΔK) in barrier models, but with clear geographic populations (Fig. B1).

COMPUTATIONAL LIMITATIONS

A neutral landscape was initiated with 14700 individuals and by generation 200 was

14131. Increasing the number of individuals on a 1050 km2 landscape resulted in intermittent failure due to limitations with computing genotypes, with 20 microsatellite loci. The mean number of alleles/locus at generations 0 and 200 was 11.0 – 9.8.

Heterozygosity remained consistent with neutral model expectations (Ho = 0.769 –

0.815). Genetic differentiation ranged from 0.0007 – 0.0017 at generations 0 and 200.

Population genetic structure at generation 200 inferred a genetic cline (Pr(X|K) = 1, ΔK =

3; Fig. B2). 171

FIGURES

Fig. B 1. Model behavior results reported for three modeled landscapes: neutral with no resistance, absolute vertical barrier, absolute horizontal barrier. Landscapes were modeled at 625 km2 with 14 animals/km2. Left to right: times-series of A) number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) population genetic pattern results from sPCA; C) STRUCTURE results. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1.

172

Fig. B 2. Computational limitations model results reported for a neutral landscape with no resistance. Landscape was modeled at 1050 km2 with 14 animals/km2. Left to right: times-series of A) number of individuals (N), observed heterozygosity (Ho), genetic differentiation (FST); B) population genetic pattern results from sPCA; C) STRUCTURE results. Because Pr(X|K) may overestimate clusters with IBD ΔK is reported, except where Pr(X|K) = 1.